<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Forem: Victor Olvera Thome (Vico)</title>
    <description>The latest articles on Forem by Victor Olvera Thome (Vico) (@vicodev).</description>
    <link>https://forem.com/vicodev</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1014732%2F64723810-2db9-4960-a54b-cf610ec8164e.png</url>
      <title>Forem: Victor Olvera Thome (Vico)</title>
      <link>https://forem.com/vicodev</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://forem.com/feed/vicodev"/>
    <language>en</language>
    <item>
      <title>Supercomputer Coatlicue: Scientific ambition or political spectacle?</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Fri, 28 Nov 2025 19:08:50 +0000</pubDate>
      <link>https://forem.com/vicodev/supercomputer-coatlicue-scientific-ambition-or-political-spectacle-4dif</link>
      <guid>https://forem.com/vicodev/supercomputer-coatlicue-scientific-ambition-or-political-spectacle-4dif</guid>
      <description>&lt;p&gt;In recent weeks, Mexican media and government officials have mentioned the alleged development of a supercomputer named &lt;strong&gt;Coatlicue&lt;/strong&gt;, presented as a national project meant to drive artificial intelligence, scientific research, and technological innovation. But beyond the announcements — what do we really know? What could we expect if it were true? And what risks would it entail, given Mexico’s recent history with large-scale public projects?&lt;/p&gt;




&lt;h3&gt;
  
  
  1️⃣ The facts: what was said and how it compares
&lt;/h3&gt;

&lt;p&gt;According to official press releases and media coverage, &lt;strong&gt;Coatlicue&lt;/strong&gt; would be a supercomputer built under Mexico’s technological sovereignty policy, with an estimated investment of &lt;strong&gt;6 billion pesos&lt;/strong&gt;. Its projected performance would reach &lt;strong&gt;314 petaflops&lt;/strong&gt;, which would — at least on paper — make it the most powerful machine in Latin America and among the world’s top ten supercomputers. That would put it above &lt;strong&gt;Leonardo&lt;/strong&gt;, the European pre-exascale supercomputer designed for scientific simulations, climate modeling, and artificial intelligence, which reaches about &lt;strong&gt;250 petaflops&lt;/strong&gt; (costing 240 million euros in 2022).&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fkz3k50zbn5m1u6vrydbp.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fkz3k50zbn5m1u6vrydbp.png" alt=" " width="800" height="436"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The government stated that Coatlicue would serve for &lt;strong&gt;scientific simulations, artificial intelligence, climate analysis, natural language processing, and predictive modeling in health, energy, and education&lt;/strong&gt;. In other words, it would be a general-purpose computing infrastructure prioritized for public and academic use.&lt;/p&gt;

&lt;p&gt;If it were to materialize, Coatlicue would surpass Mexico’s current supercomputer &lt;strong&gt;Miztli&lt;/strong&gt;, located at UNAM, which reaches around &lt;strong&gt;1.3 petaflops&lt;/strong&gt;, and even &lt;strong&gt;Santos Dumont (Brazil)&lt;/strong&gt;, Latin America’s current leader with roughly &lt;strong&gt;15 petaflops&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;For global perspective, the most powerful computer in the world — &lt;strong&gt;Frontier&lt;/strong&gt;, at Oak Ridge National Laboratory (U.S.) — exceeds &lt;strong&gt;1.1 exaflops&lt;/strong&gt; (1,100 petaflops). That means Coatlicue, even at its best, would have roughly a quarter of Frontier’s capacity — enough to position Mexico competitively and mark a significant leap for the country and the region.&lt;/p&gt;

&lt;p&gt;However, &lt;strong&gt;no verifiable data&lt;/strong&gt; has been provided about its location, hardware supplier, or related contracts. Everything so far points to a political announcement rather than a tangible project under construction.&lt;/p&gt;




&lt;h3&gt;
  
  
  2️⃣ First &lt;em&gt;What if&lt;/em&gt;: What if Coatlicue were real and fulfilled its promise?
&lt;/h3&gt;

&lt;p&gt;Let’s assume Coatlicue truly exists — and works. In that case, the project could mark a turning point for Mexican science. Not just for the infrastructure, but for what it would mean in terms of collaboration, talent, and technological independence.&lt;/p&gt;

&lt;p&gt;A supercomputer of that magnitude would allow universities and research centers to run AI models that are currently unfeasible due to hardware limitations. It could, for example:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Train &lt;strong&gt;language models in Spanish and indigenous languages&lt;/strong&gt; without relying on foreign infrastructure.&lt;/li&gt;
&lt;li&gt;Perform &lt;strong&gt;climate simulations&lt;/strong&gt; with greater precision, supporting environmental policies and disaster prevention.&lt;/li&gt;
&lt;li&gt;Accelerate &lt;strong&gt;medical research&lt;/strong&gt; through rapid genomic and pharmacological analysis.&lt;/li&gt;
&lt;li&gt;Support &lt;strong&gt;AI applications in public education&lt;/strong&gt;, adaptive learning, and large-scale data analytics.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Beyond its computing power, Coatlicue could become a &lt;strong&gt;symbol of national pride&lt;/strong&gt;, proving that Mexico can develop world-class scientific infrastructure. On an industrial level, it could attract innovation projects that today migrate to countries like Brazil, Chile, or even Spain.&lt;/p&gt;

&lt;p&gt;If managed with transparency and long-term vision, it could also &lt;strong&gt;reduce Mexico’s digital divide&lt;/strong&gt; by providing access to state universities and startups that currently lack advanced computing capabilities.&lt;/p&gt;

&lt;p&gt;In short, Coatlicue could be the start of a &lt;strong&gt;quiet revolution&lt;/strong&gt;: an infrastructure designed not only to calculate, but to inspire.&lt;/p&gt;




&lt;h3&gt;
  
  
  3️⃣ Second &lt;em&gt;What if&lt;/em&gt;: the darker scenario
&lt;/h3&gt;

&lt;p&gt;Now let’s look at the other side of the coin. What if Coatlicue existed but became yet another grand promise lost to opacity, overcosts, or abandonment?&lt;/p&gt;

&lt;p&gt;The recent track record of Mexico’s so-called “Fourth Transformation” megaprojects is not encouraging. The &lt;strong&gt;Maya Train&lt;/strong&gt;, the &lt;strong&gt;Dos Bocas Refinery&lt;/strong&gt;, and the &lt;strong&gt;Felipe Ángeles Airport&lt;/strong&gt; have all faced criticism for budget overruns, lack of planning, and underwhelming outcomes. The recurring pattern: &lt;strong&gt;a lack of transparency&lt;/strong&gt; and &lt;strong&gt;the politicization of technology&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Within that context, Coatlicue could become just another emblematic project — scientific on paper, political in practice.&lt;/p&gt;

&lt;p&gt;There’s also a more concerning risk: the &lt;strong&gt;misuse of technological infrastructure for surveillance or censorship&lt;/strong&gt;. Mexico has a documented history of government espionage, such as the &lt;strong&gt;Pegasus&lt;/strong&gt; case, where digital tools were used to spy on journalists, activists, and opposition figures. A supercomputer capable of processing vast amounts of data could — in the worst-case scenario — be used to power control systems or manipulate information.&lt;/p&gt;

&lt;p&gt;Another issue is &lt;strong&gt;bureaucratic inefficiency&lt;/strong&gt;. If Coatlicue were managed by agencies with limited technical expertise or without clear access policies, it could end up underutilized or monopolized by politically favored institutions.&lt;/p&gt;

&lt;p&gt;There are also financial red flags. When compared with similar projects worldwide, the numbers don’t quite add up. For instance, &lt;strong&gt;Alps&lt;/strong&gt;, the Swiss supercomputer inaugurated in September 2024, cost &lt;strong&gt;100 million Swiss francs&lt;/strong&gt; (around &lt;strong&gt;2.27 billion pesos&lt;/strong&gt;) and reaches &lt;strong&gt;435 petaflops&lt;/strong&gt; — roughly &lt;strong&gt;30% more performance than Coatlicue’s projection&lt;/strong&gt;, at &lt;strong&gt;less than half the cost&lt;/strong&gt; (6 billion pesos).&lt;/p&gt;

&lt;p&gt;The likely result: a powerful but inefficient machine — heavy on symbolism, light on substance.&lt;/p&gt;




&lt;h3&gt;
  
  
  4️⃣ Conclusions: between hope and skepticism
&lt;/h3&gt;

&lt;p&gt;Thinking about Coatlicue sparks mixed feelings. On one hand, it’s inspiring to imagine a Mexican supercomputer serving science — training AI models in Spanish, running climate simulations, or tackling complex public health problems. It would be a statement of ambition and technological sovereignty.&lt;/p&gt;

&lt;p&gt;On the other hand, experience tells us that Mexico’s grand public projects often stumble on poor planning, corruption, and political protagonism. Science doesn’t advance through speeches — it thrives on continuity, transparency, and collaboration.&lt;/p&gt;

&lt;p&gt;Ultimately, Coatlicue remains more of a &lt;strong&gt;powerful idea&lt;/strong&gt; than a tangible reality. And like all powerful ideas, it can be used to build — or to control.&lt;/p&gt;

&lt;p&gt;If this supercomputer ever comes to life, let’s hope it does so to enlighten knowledge, not to obscure it.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;References:&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Presidencia de la República. (2025, November 26). &lt;em&gt;Presidenta Claudia Sheinbaum presenta “Coatlicue”, la supercomputadora del pueblo de México y la más poderosa de América Latina.&lt;/em&gt; Gobierno de México. &lt;a href="https://www.gob.mx/presidencia/prensa/presidenta-claudia-sheinbaum-presenta-coatlicue-la-supercomputadora-del-pueblo-de-mexico-y-la-mas-poderosa-de-america-latina" rel="noopener noreferrer"&gt;https://www.gob.mx/presidencia/prensa/presidenta-claudia-sheinbaum-presenta-coatlicue-la-supercomputadora-del-pueblo-de-mexico-y-la-mas-poderosa-de-america-latina&lt;/a&gt;&lt;br&gt;&lt;br&gt;
TOP500. (2025). &lt;em&gt;HPCG Benchmark — November 2025 list.&lt;/em&gt; &lt;a href="https://top500.org/lists/hpcg/2025/11/" rel="noopener noreferrer"&gt;https://top500.org/lists/hpcg/2025/11/&lt;/a&gt;&lt;br&gt;&lt;br&gt;
The Next Platform. (2024, September 23). &lt;em&gt;A GPU upgrade for “Leonardo” supercomputer but not a budget upgrade.&lt;/em&gt; &lt;a href="https://www.nextplatform.com/2024/09/23/a-gpu-upgrade-for-leonardo-supercomputer-but-not-a-budget-upgrade/" rel="noopener noreferrer"&gt;https://www.nextplatform.com/2024/09/23/a-gpu-upgrade-for-leonardo-supercomputer-but-not-a-budget-upgrade/&lt;/a&gt;&lt;br&gt;&lt;br&gt;
Swissinfo.ch. (2024, October 3). &lt;em&gt;How Switzerland’s Alps supercomputer aims to advance AI.&lt;/em&gt; &lt;a href="https://www.swissinfo.ch/eng/science/how-switzerlands-alps-supercomputer-aims-to-advance-ai/87659724" rel="noopener noreferrer"&gt;https://www.swissinfo.ch/eng/science/how-switzerlands-alps-supercomputer-aims-to-advance-ai/87659724&lt;/a&gt;&lt;br&gt;&lt;br&gt;
ETH Zurich. (2024, September 14). &lt;em&gt;New research infrastructure: “Alps” supercomputer inaugurated.&lt;/em&gt; ETH News. &lt;a href="https://www.ethz.ch/en/news-and-events/eth-news/news/2024/09/press-release-new-research-infrastructure-alps-supercomputer-inaugurated.html" rel="noopener noreferrer"&gt;https://www.ethz.ch/en/news-and-events/eth-news/news/2024/09/press-release-new-research-infrastructure-alps-supercomputer-inaugurated.html&lt;/a&gt;  &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;coatlicue&lt;/code&gt;, &lt;code&gt;computerscience&lt;/code&gt;, &lt;code&gt;mexico&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>coatlicue</category>
      <category>computerscience</category>
      <category>mexico</category>
    </item>
    <item>
      <title>Supercomputadora Coatlicue: ¿ambición científica o espectáculo político?</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Fri, 28 Nov 2025 19:05:55 +0000</pubDate>
      <link>https://forem.com/vicodev/coatlicue-y-si-la-supercomputadora-mexicana-realmente-existiera-2bl7</link>
      <guid>https://forem.com/vicodev/coatlicue-y-si-la-supercomputadora-mexicana-realmente-existiera-2bl7</guid>
      <description>&lt;p&gt;Durante las últimas semanas, medios y funcionarios mexicanos mencionaron el supuesto desarrollo de una supercomputadora llamada &lt;strong&gt;Coatlicue&lt;/strong&gt;, presentada como un proyecto nacional que impulsaría la inteligencia artificial, la investigación científica y la innovación tecnológica. Pero, más allá de los anuncios, ¿qué sabemos realmente?, ¿qué podríamos esperar si fuera cierta?, y ¿qué riesgos implicaría considerando la historia reciente de los grandes proyectos públicos en México? &lt;/p&gt;




&lt;h3&gt;
  
  
  1️⃣ Los hechos: lo que se dijo y lo que podemos comparar
&lt;/h3&gt;

&lt;p&gt;De acuerdo con la información difundida en comunicados oficiales y notas de prensa, &lt;strong&gt;Coatlicue&lt;/strong&gt; sería una supercomputadora instalada en el marco de la política de soberanía tecnológica, con una inversión estimada de &lt;strong&gt;6 mil millones de pesos&lt;/strong&gt;. Su potencia proyectada sería de &lt;strong&gt;314 petaflops&lt;/strong&gt;, lo que la colocaría —al menos en papel— como la máquina más potente de América Latina y en el top 10 de las supercomputadoras a nivel mundial, por encima del 8vo sitio que lo ocupa Leonardo, supercomputadora europea pre-exascala, diseñada para simulaciones científicas, clima e inteligencia artificial, que alcanza cerca de 250 petaflops (240 millones de Euros, 2022). &lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fkz3k50zbn5m1u6vrydbp.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fkz3k50zbn5m1u6vrydbp.png" alt=" " width="800" height="436"&gt;&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;El gobierno habría mencionado que Coatlicue serviría para tareas de &lt;strong&gt;simulación científica, inteligencia artificial, análisis climático, procesamiento de lenguaje natural y desarrollo de modelos predictivos para salud, energía y educación&lt;/strong&gt;. En otras palabras, sería una infraestructura de cómputo de propósito general, con prioridad para proyectos públicos y académicos. &lt;/p&gt;

&lt;p&gt;Si esto se materializara, Coatlicue superaría a la actual supercomputadora mexicana &lt;strong&gt;Miztli&lt;/strong&gt;, ubicada en la UNAM, que ronda los &lt;strong&gt;1.3 petaflops&lt;/strong&gt;, y a la líder latinoamericana &lt;strong&gt;Santos Dumont (Brasil)&lt;/strong&gt; con cerca de &lt;strong&gt;15 petaflops&lt;/strong&gt; de capacidad. &lt;/p&gt;

&lt;p&gt;Por ponerlo en perspectiva global, la máquina más poderosa del mundo —&lt;strong&gt;Frontier&lt;/strong&gt;, en Oak Ridge National Laboratory (EE.UU.)— supera los &lt;strong&gt;1.1 exaflops&lt;/strong&gt; (1,100 petaflops), lo que significa que Coatlicue, en su mejor versión, tendría algo así como una cuarta parte de potencia. Sería entonces suficiente para colocar a México en un nivel competitivo y representar un salto significativo respecto a la capacidad actual del país y el continente. &lt;/p&gt;

&lt;p&gt;Hasta ahora, sin embargo, &lt;strong&gt;no existen datos verificables&lt;/strong&gt; sobre su ubicación, su proveedor de hardware ni los contratos asociados. Todo apunta más a un anuncio político que a un proyecto en ejecución tangible. &lt;/p&gt;




&lt;h3&gt;
  
  
  2️⃣ Primer &lt;em&gt;What if&lt;/em&gt;: ¿y si Coatlicue fuera real y cumpliera su promesa?
&lt;/h3&gt;

&lt;p&gt;Supongamos que Coatlicue efectivamente existe, y que funciona. En ese escenario, el proyecto podría marcar un antes y un después para la ciencia mexicana. No solo por la infraestructura, sino por lo que implicaría en términos de colaboración, talento e independencia tecnológica. &lt;/p&gt;

&lt;p&gt;Una supercomputadora de esa magnitud permitiría a universidades y centros de investigación ejecutar modelos de IA que hoy son inviables por limitaciones de hardware. Podría, por ejemplo: &lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Entrenar &lt;strong&gt;modelos de lenguaje en español y lenguas indígenas&lt;/strong&gt; sin depender de recursos extranjeros. &lt;/li&gt;
&lt;li&gt;Procesar &lt;strong&gt;simulaciones climáticas&lt;/strong&gt; con mayor precisión, apoyando políticas ambientales y de prevención de desastres. &lt;/li&gt;
&lt;li&gt;Impulsar &lt;strong&gt;investigación médica&lt;/strong&gt; mediante análisis genómicos y farmacológicos en tiempo récord. &lt;/li&gt;
&lt;li&gt;Servir de soporte para &lt;strong&gt;aplicaciones de IA en educación pública&lt;/strong&gt;, personalización de contenidos y análisis de datos de rendimiento académico.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Además, Coatlicue podría convertirse en un símbolo de orgullo nacional, demostrando que México puede desarrollar infraestructura científica de primer nivel. A nivel industrial, permitiría al país atraer proyectos de innovación que hoy migran hacia otros centros tecnológicos, como Brasil, Chile o incluso España. &lt;/p&gt;

&lt;p&gt;Si se gestionara con transparencia y visión, también podría &lt;strong&gt;reducir la brecha digital&lt;/strong&gt; al ofrecer recursos a universidades estatales y startups tecnológicas con acceso limitado al cómputo avanzado. &lt;/p&gt;

&lt;p&gt;En otras palabras, Coatlicue podría ser el punto de partida de una &lt;strong&gt;revolución silenciosa&lt;/strong&gt;: una infraestructura que no solo calcule, sino que inspire.&lt;/p&gt;




&lt;h3&gt;
  
  
  3️⃣ Segundo &lt;em&gt;What if&lt;/em&gt;: la parte perversa
&lt;/h3&gt;

&lt;p&gt;Ahora, vayamos al otro lado del espejo. ¿Y si Coatlicue existiera, pero se convirtiera en otro de esos proyectos que empiezan con promesas grandiosas y terminan en opacidad, sobrecostos o abandono? &lt;/p&gt;

&lt;p&gt;La historia reciente de los megaproyectos de la llamada 4T no es precisamente alentadora. El &lt;strong&gt;Tren Maya&lt;/strong&gt;, la &lt;strong&gt;Refinería Dos Bocas&lt;/strong&gt; o el &lt;strong&gt;Aeropuerto Felipe Ángeles&lt;/strong&gt; han enfrentado críticas por sobrecostos, falta de planeación y resultados por debajo de lo esperado. La constante ha sido la &lt;strong&gt;falta de transparencia&lt;/strong&gt; y la &lt;strong&gt;politización de la tecnología&lt;/strong&gt;. &lt;/p&gt;

&lt;p&gt;En este contexto, Coatlicue podría correr el riesgo de convertirse en un proyecto emblemático más, con objetivos científicos en el papel, pero control político en la práctica. &lt;/p&gt;

&lt;p&gt;Además, existe un tema más delicado: el &lt;strong&gt;uso indebido de la infraestructura tecnológica para vigilancia o censura&lt;/strong&gt;. México tiene un historial documentado de espionaje gubernamental, como el caso de &lt;strong&gt;Pegasus&lt;/strong&gt;, donde se utilizaron herramientas digitales para espiar a periodistas, activistas y opositores. Una supercomputadora con acceso a grandes volúmenes de datos podría —en el peor de los escenarios— alimentar sistemas de control y manipulación de información. &lt;/p&gt;

&lt;p&gt;Otro riesgo es la &lt;strong&gt;ineficiencia burocrática&lt;/strong&gt;. Si Coatlicue fuera administrada por dependencias sin experiencia técnica o sin políticas de acceso claras, podría terminar infrautilizada o monopolizada por unas cuantas instituciones cercanas al poder. &lt;/p&gt;

&lt;p&gt;También existen indicadores que invitan a dudar desde ahora, especialmente si comparamos con proyectos ya concluidos en el mundo. Alps, la supercomputadora suiza inaugurada en septiembre de 2024, tuvo un costo de 100 millones de francos suizos (aprox. 2,276 millones de pesos) y alcanza 435 petaflops de rendimiento, alrededor de &lt;strong&gt;30% más que lo proyectado para Coatlicue&lt;/strong&gt;, que cuenta con un presupuesto de 6,000 millones de pesos, es decir, &lt;strong&gt;casi el triple del costo de Alps&lt;/strong&gt;. &lt;/p&gt;

&lt;p&gt;El resultado: una máquina poderosa, pero sin impacto real. &lt;/p&gt;




&lt;h3&gt;
  
  
  4️⃣ Conclusiones: entre la esperanza y la desconfianza
&lt;/h3&gt;

&lt;p&gt;Pensar en Coatlicue despierta emociones encontradas. Por un lado, es inspirador imaginar una supercomputadora mexicana al servicio de la ciencia, capaz de entrenar modelos de IA en español, impulsar investigación climática o ayudar a resolver problemas complejos de salud pública. Sería una declaración de ambición y soberanía tecnológica. &lt;/p&gt;

&lt;p&gt;Pero por otro lado, la experiencia nos enseña que los grandes proyectos públicos en México suelen tropezar con la falta de planeación, la corrupción y el protagonismo político. La ciencia no avanza con discursos, sino con continuidad, transparencia y comunidad. &lt;/p&gt;

&lt;p&gt;A fin de cuentas, Coatlicue es hoy más una &lt;strong&gt;idea poderosa&lt;/strong&gt; que una realidad tangible. Y como toda idea poderosa, puede ser usada para construir o para controlar. &lt;/p&gt;

&lt;p&gt;Si algún día esta supercomputadora se enciende, esperemos que lo haga para iluminar el conocimiento, no para oscurecerlo. &lt;/p&gt;




&lt;p&gt;Presidencia de la República. (2025, 26 de noviembre). Presidenta Claudia Sheinbaum presenta “Coatlicue”, la supercomputadora del pueblo de México y la más poderosa de América Latina. Gobierno de México. &lt;a href="https://www.gob.mx/presidencia/prensa/presidenta-claudia-sheinbaum-presenta-coatlicue-la-supercomputadora-del-pueblo-de-mexico-y-la-mas-poderosa-de-america-latina" rel="noopener noreferrer"&gt;https://www.gob.mx/presidencia/prensa/presidenta-claudia-sheinbaum-presenta-coatlicue-la-supercomputadora-del-pueblo-de-mexico-y-la-mas-poderosa-de-america-latina&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;TOP500. (2025). HPCG Benchmark — November 2025 list. &lt;a href="https://top500.org/lists/hpcg/2025/11/" rel="noopener noreferrer"&gt;https://top500.org/lists/hpcg/2025/11/&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;The Next Platform. (2024, 23 de septiembre). A GPU upgrade for “Leonardo” supercomputer but not a budget upgrade. The Next Platform. &lt;a href="https://www.nextplatform.com/2024/09/23/a-gpu-upgrade-for-leonardo-supercomputer-but-not-a-budget-upgrade/" rel="noopener noreferrer"&gt;https://www.nextplatform.com/2024/09/23/a-gpu-upgrade-for-leonardo-supercomputer-but-not-a-budget-upgrade/&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;Swissinfo.ch. (2024, 3 de octubre). How Switzerland’s Alps supercomputer aims to advance AI. &lt;a href="https://www.swissinfo.ch/eng/science/how-switzerlands-alps-supercomputer-aims-to-advance-ai/87659724" rel="noopener noreferrer"&gt;https://www.swissinfo.ch/eng/science/how-switzerlands-alps-supercomputer-aims-to-advance-ai/87659724&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;ETH Zurich. (2024, 14 de septiembre). New research infrastructure: "Alps" supercomputer inaugurated. ETH News. &lt;a href="https://www.ethz.ch/en/news-and-events/eth-news/news/2024/09/press-release-new-research-infrastructure-alps-supercomputer-inaugurated.html" rel="noopener noreferrer"&gt;https://www.ethz.ch/en/news-and-events/eth-news/news/2024/09/press-release-new-research-infrastructure-alps-supercomputer-inaugurated.html&lt;/a&gt; &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; ai, Coatlicue, ComputerScience, Mexico&lt;/p&gt;

</description>
      <category>ai</category>
      <category>coatlicue</category>
      <category>mexico</category>
      <category>computerscience</category>
    </item>
    <item>
      <title>El conocimiento lingüístico en NLP: el puente entre la sintaxis y la semántica</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Wed, 26 Nov 2025 01:10:49 +0000</pubDate>
      <link>https://forem.com/vicodev/el-conocimiento-linguistico-en-nlp-el-puente-entre-la-sintaxis-y-la-semantica-2m8a</link>
      <guid>https://forem.com/vicodev/el-conocimiento-linguistico-en-nlp-el-puente-entre-la-sintaxis-y-la-semantica-2m8a</guid>
      <description>&lt;p&gt;La inteligencia artificial moderna ha avanzado enormemente en el procesamiento de lenguaje natural (NLP), pero sigue enfrentándose a una pregunta esencial: ¿entienden las máquinas el lenguaje o simplemente lo imitan? Aquí entra en juego el &lt;strong&gt;conocimiento lingüístico&lt;/strong&gt;, el conjunto de reglas, estructuras y significados que los humanos utilizamos para comunicarnos de manera coherente.&lt;/p&gt;




&lt;p&gt;Durante décadas, el NLP se apoyó en la lingüística tradicional. Los sistemas estaban construidos sobre gramáticas, parsers y reglas sintácticas, reflejando una comprensión estructural del idioma. Sin embargo, con la llegada del aprendizaje profundo, este enfoque dio paso a modelos basados en datos masivos. Las redes neuronales comenzaron a inferir patrones estadísticos, sin depender explícitamente de la teoría lingüística.&lt;/p&gt;




&lt;p&gt;Hoy, modelos como &lt;strong&gt;BERT&lt;/strong&gt;, &lt;strong&gt;GPT&lt;/strong&gt; o &lt;strong&gt;Gemini&lt;/strong&gt; parecen haber aprendido a representar significado. Sin embargo, lo hacen de forma implícita: aprenden asociaciones entre palabras, contextos y relaciones gramaticales a través de billones de ejemplos, pero sin una noción formal de sintaxis o semántica. Aun así, muchos investigadores coinciden en que estas redes aprenden &lt;em&gt;aproximaciones lingüísticas emergentes&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;Esto plantea una paradoja fascinante: los modelos de IA pueden superar a los humanos en tareas lingüísticas específicas, sin realmente “entender” el lenguaje. El conocimiento lingüístico humano no solo estructura oraciones; también modela ambigüedad, ironía, implicaturas y contexto cultural.&lt;/p&gt;




&lt;p&gt;El valor de integrar lingüística y deep learning no está en reemplazar datos por reglas, sino en &lt;strong&gt;equilibrar conocimiento y aprendizaje&lt;/strong&gt;. Incluir información sintáctica o semántica explícita puede hacer que los modelos sean más interpretables, más eficientes y menos dependientes de grandes volúmenes de datos. Por ejemplo, los modelos híbridos que combinan embeddings con información gramatical han demostrado mejoras en comprensión de texto y razonamiento lógico.&lt;/p&gt;




&lt;p&gt;Los desafíos son amplios. La diversidad lingüística introduce sesgos culturales y estructurales que dificultan la generalización. Además, los modelos suelen reflejar las asimetrías del lenguaje dominante en sus datos de entrenamiento. Incorporar conocimiento lingüístico explícito también requiere colaboración interdisciplinaria: lingüistas, computólogos y especialistas en IA trabajando juntos.&lt;/p&gt;




&lt;p&gt;A futuro, la sinergia entre lingüística e inteligencia artificial podría conducir a sistemas más conscientes del contexto humano. En lugar de reemplazar el conocimiento lingüístico, la IA podría &lt;strong&gt;redescubrirlo desde otra perspectiva&lt;/strong&gt;, creando un puente entre el razonamiento estadístico y el entendimiento semántico.&lt;/p&gt;




&lt;p&gt;Checklist final:&lt;br&gt;
✅ Comprender el papel del conocimiento lingüístico en el NLP.&lt;br&gt;&lt;br&gt;
✅ Reconocer los límites del aprendizaje puramente estadístico.&lt;br&gt;&lt;br&gt;
✅ Explorar formas de integrar teoría lingüística en modelos de IA modernos.  &lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;nlp&lt;/code&gt;, &lt;code&gt;linguistics&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>nlp</category>
      <category>linguistics</category>
      <category>programming</category>
    </item>
    <item>
      <title>Linguistic Knowledge in NLP: bridging syntax and semantics</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Wed, 26 Nov 2025 01:10:19 +0000</pubDate>
      <link>https://forem.com/vicodev/linguistic-knowledge-in-nlp-bridging-syntax-and-semantics-11a7</link>
      <guid>https://forem.com/vicodev/linguistic-knowledge-in-nlp-bridging-syntax-and-semantics-11a7</guid>
      <description>&lt;p&gt;Modern artificial intelligence has made tremendous progress in natural language processing (NLP), yet it still faces a profound question: do machines truly understand language, or are they simply mimicking it? This is where &lt;strong&gt;linguistic knowledge&lt;/strong&gt; comes into play — the set of rules, structures, and meanings humans use to communicate coherently.&lt;/p&gt;

&lt;p&gt;For decades, NLP was grounded in traditional linguistics. Early systems relied on grammars, parsers, and syntactic rules, reflecting a structured understanding of language. However, with the rise of deep learning, this approach gave way to data-driven models. Neural networks began to infer statistical patterns, bypassing explicit linguistic theory.&lt;/p&gt;

&lt;p&gt;Today, models like &lt;strong&gt;BERT&lt;/strong&gt;, &lt;strong&gt;GPT&lt;/strong&gt;, and &lt;strong&gt;Gemini&lt;/strong&gt; seem to grasp meaning. Yet they do so implicitly — by learning associations between words, contexts, and grammatical relations from trillions of examples, without any formal notion of syntax or semantics. Still, many researchers agree that these networks learn &lt;em&gt;emergent linguistic approximations&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;This leads to a fascinating paradox: AI models can outperform humans in linguistic tasks without truly “understanding” language. Human linguistic knowledge not only structures sentences but also models ambiguity, irony, implicature, and cultural context.&lt;/p&gt;

&lt;p&gt;The key to integrating linguistics and deep learning isn’t replacing data with rules — it’s &lt;strong&gt;balancing knowledge and learning&lt;/strong&gt;. Including explicit syntactic or semantic information can make models more interpretable, efficient, and less dependent on massive datasets. Hybrid models that combine embeddings with grammatical knowledge have shown improvements in text comprehension and logical reasoning.&lt;/p&gt;

&lt;p&gt;However, challenges remain. Linguistic diversity introduces cultural and structural biases that complicate generalization. Models tend to reflect the asymmetries of dominant languages in their training data. Integrating linguistic knowledge also demands interdisciplinary collaboration among linguists, computer scientists, and AI researchers.&lt;/p&gt;

&lt;p&gt;In the future, the synergy between linguistics and artificial intelligence could lead to systems more aware of human context. Instead of replacing linguistic knowledge, AI might &lt;strong&gt;rediscover it through new perspectives&lt;/strong&gt;, building a bridge between statistical reasoning and semantic understanding.&lt;/p&gt;

&lt;p&gt;Final checklist:&lt;br&gt;
✅ Understand the role of linguistic knowledge in NLP.&lt;br&gt;&lt;br&gt;
✅ Recognize the limits of purely statistical learning.&lt;br&gt;&lt;br&gt;
✅ Explore ways to integrate linguistic theory into modern AI models.  &lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;nlp&lt;/code&gt;, &lt;code&gt;linguistics&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>nlp</category>
      <category>linguistics</category>
      <category>programming</category>
    </item>
    <item>
      <title>Gemini 3: the multimodal leap redefining Google’s artificial intelligence</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Thu, 20 Nov 2025 23:11:46 +0000</pubDate>
      <link>https://forem.com/vicodev/gemini-3-the-multimodal-leap-redefining-googles-artificial-intelligence-3fa0</link>
      <guid>https://forem.com/vicodev/gemini-3-the-multimodal-leap-redefining-googles-artificial-intelligence-3fa0</guid>
      <description>&lt;p&gt;Artificial intelligence is entering its most transformative era, and Google isn’t staying behind. With &lt;strong&gt;Gemini 3&lt;/strong&gt;, the company introduces a multimodal model that marks a leap toward AI capable of understanding, combining, and generating across formats — text, image, audio, video, and code. More than an update, Gemini 3 represents a new paradigm of cognitive integration between humans and machines.&lt;/p&gt;




&lt;p&gt;The evolution of Gemini has been steady: from the first language-focused model, to Gemini 2, which added visual and contextual understanding. &lt;strong&gt;Gemini 3 merges all these abilities&lt;/strong&gt; into an architecture optimized for complex reasoning, multimodal synthesis, and continuous dialogue.&lt;/p&gt;

&lt;p&gt;The model can retain context through long interactions, analyze images and code simultaneously, and produce content with both narrative and technical coherence. This makes it ideal for developers, researchers, and creators exploring ML and AI applications.&lt;/p&gt;




&lt;p&gt;Unlike previous generations, Gemini 3 doesn’t just respond. It learns from the flow of conversation, adjusts tone, and can “reason” through technical decisions. This enables new use cases — from intelligent programming assistants to personalized tutors or creative partners.&lt;/p&gt;

&lt;p&gt;Its training across diverse datasets allows it to recognize cross-domain patterns — for example, understanding a software architecture diagram and converting it into executable code, or describing a financial chart in natural language.&lt;/p&gt;




&lt;p&gt;Why it matters:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Cognitive scalability:&lt;/strong&gt; Gemini 3 brings us closer to systems that truly understand context beyond words.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Seamless integration:&lt;/strong&gt; its multimodal approach lets developers work across complex data types with a single model.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Technological accessibility:&lt;/strong&gt; integration into tools like Colab and Vertex AI democratizes access to advanced AI.
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;Gemini 3 is no longer just a model — it’s a platform adapting to every task. Google aims to bridge users and AI more naturally, enhancing productivity, creativity, and research through intelligence that feels intuitive and human.&lt;/p&gt;




&lt;p&gt;Final checklist:&lt;br&gt;
✅ Understand what makes Gemini 3 unique.&lt;br&gt;&lt;br&gt;
✅ Explore its multimodal potential in real projects.&lt;br&gt;&lt;br&gt;
✅ Reflect on the ethical and social implications of integrated AI.  &lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;google&lt;/code&gt;, &lt;code&gt;gemini&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>google</category>
      <category>gemini</category>
      <category>programming</category>
    </item>
    <item>
      <title>Gemini 3: el salto multimodal que redefine la inteligencia artificial de Google</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Thu, 20 Nov 2025 23:10:57 +0000</pubDate>
      <link>https://forem.com/vicodev/gemini-3-el-salto-multimodal-que-redefine-la-inteligencia-artificial-de-google-1nn0</link>
      <guid>https://forem.com/vicodev/gemini-3-el-salto-multimodal-que-redefine-la-inteligencia-artificial-de-google-1nn0</guid>
      <description>&lt;p&gt;La inteligencia artificial está atravesando su momento más transformador, y Google no se queda atrás. Con &lt;strong&gt;Gemini 3&lt;/strong&gt;, su nuevo modelo multimodal, la compañía da un salto cualitativo hacia una IA que entiende, combina y genera información en múltiples formatos: texto, imagen, audio, video y código. Más que una simple actualización, Gemini 3 representa un nuevo paradigma de integración cognitiva entre humano y máquina.&lt;/p&gt;




&lt;p&gt;El desarrollo de Gemini ha sido progresivo: desde el primer modelo enfocado en lenguaje natural, hasta Gemini 2 que amplió la comprensión visual y contextual. &lt;strong&gt;Gemini 3 fusiona todas esas capacidades&lt;/strong&gt; en una arquitectura optimizada para el razonamiento complejo, la síntesis multimodal y el diálogo continuo.&lt;/p&gt;

&lt;p&gt;El modelo puede mantener contexto durante interacciones prolongadas, analizar imágenes y código simultáneamente, y generar contenido con coherencia narrativa y técnica. Esto lo convierte en una herramienta ideal tanto para desarrolladores como para investigadores en ML/IA.&lt;/p&gt;




&lt;p&gt;A diferencia de generaciones anteriores, Gemini 3 no se limita a responder. Aprende del flujo de la conversación, ajusta su tono y puede “razonar” sobre decisiones técnicas. Esto abre nuevas posibilidades para aplicaciones como asistentes de programación, tutores personalizados o herramientas creativas.&lt;/p&gt;

&lt;p&gt;La potencia de su entrenamiento en datos diversos le permite reconocer patrones entre dominios distintos. Por ejemplo, comprender un diagrama de arquitectura y transformarlo en código ejecutable, o explicar un gráfico financiero en lenguaje natural.&lt;/p&gt;




&lt;p&gt;Por qué importa:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Escalabilidad cognitiva:&lt;/strong&gt; Gemini 3 representa un paso hacia sistemas que comprenden el contexto más allá de las palabras.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Integración fluida:&lt;/strong&gt; su enfoque multimodal permite que los desarrolladores trabajen con datos complejos sin necesidad de múltiples modelos.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accesibilidad tecnológica:&lt;/strong&gt; al integrarse en productos como Colab o Vertex AI, democratiza el acceso a IA avanzada.
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;Gemini 3 ya no es solo un modelo, sino una plataforma que se adapta a cada tarea. Google busca con él acercar la IA al usuario final sin fricciones, impulsando la productividad, la creatividad y la investigación.&lt;/p&gt;




&lt;p&gt;Checklist final:&lt;br&gt;
✅ Comprender qué hace único a Gemini 3.&lt;br&gt;&lt;br&gt;
✅ Explorar su potencial multimodal en proyectos reales.&lt;br&gt;&lt;br&gt;
✅ Reflexionar sobre el impacto ético y social de una IA más integrada.  &lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;google&lt;/code&gt;, &lt;code&gt;gemini&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>google</category>
      <category>gemini</category>
      <category>programming</category>
    </item>
    <item>
      <title>Antigravity IDE: the next generation of AI-powered code editors</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Tue, 18 Nov 2025 20:13:20 +0000</pubDate>
      <link>https://forem.com/vicodev/antigravity-ide-the-next-generation-of-ai-powered-code-editors-241h</link>
      <guid>https://forem.com/vicodev/antigravity-ide-the-next-generation-of-ai-powered-code-editors-241h</guid>
      <description>&lt;h1&gt;
  
  
  Antigravity IDE: the next generation of AI-powered code editors
&lt;/h1&gt;

&lt;p&gt;Artificial intelligence has transformed how we write, learn, and now, how we code. If VS Code symbolized the modern era of development, &lt;strong&gt;Antigravity IDE&lt;/strong&gt; represents the next frontier: an environment capable of understanding the developer, anticipating intentions, and collaborating proactively in software creation.&lt;/p&gt;




&lt;h2&gt;
  
  
  From VS Code to Antigravity
&lt;/h2&gt;

&lt;p&gt;The concept of &lt;em&gt;Antigravity&lt;/em&gt; doesn’t come from an actual Google tool but from a metaphor — the idea of rising above current limitations. In Python, &lt;code&gt;import antigravity&lt;/code&gt; was a humorous nod to the absurd and brilliant nature of programming. In this new context, Antigravity becomes a &lt;strong&gt;vision of AI-augmented development&lt;/strong&gt;, where the editor stops being a passive canvas and evolves into an intelligent partner.&lt;/p&gt;

&lt;p&gt;Antigravity IDE would merge the flexibility of VS Code with the power of language models, integrating systems that learn from user flow and personalize every experience. This hypothetical Google fork would be more than just a tool: &lt;strong&gt;an autonomous copilot&lt;/strong&gt;, able to grasp context, suggest solutions, and correct errors in real time.&lt;/p&gt;




&lt;h2&gt;
  
  
  AI as a conscious copilot
&lt;/h2&gt;

&lt;p&gt;An IDE with deep learning capabilities could analyze a programmer’s behavior and adapt its interface and functions based on habits. For instance, prioritizing commands or suggesting structures derived from previous projects.&lt;/p&gt;

&lt;p&gt;The system would be intelligent enough to interpret intentions, spot code ambiguities, and propose alternatives without constant user intervention. This adaptability breaks today’s “autocomplete” paradigm and opens the door to &lt;strong&gt;cognitive collaboration&lt;/strong&gt;, where AI learns from the developer as much as the developer learns from it.&lt;/p&gt;

&lt;p&gt;Imagine an environment that knows when you’re exploring an idea versus optimizing performance — distinguishing between a logical error and an intentional experiment. That’s the essence of Antigravity IDE: a workspace where AI &lt;strong&gt;doesn’t replace the human&lt;/strong&gt;, but amplifies creativity and efficiency.&lt;/p&gt;




&lt;h2&gt;
  
  
  The value of an intelligent IDE
&lt;/h2&gt;

&lt;p&gt;The shift from traditional editors to AI-powered environments isn’t merely technical — it’s cultural. It redefines the relationship between creator and tool, turning the editor into a cognitive ally.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Frictionless efficiency:&lt;/strong&gt; repetitive tasks automated, more focus on logic.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Continuous learning:&lt;/strong&gt; the environment evolves alongside the programmer’s style.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open innovation:&lt;/strong&gt; collaboration between AI, community, and open-source code.
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The future of programming isn’t just about writing code faster — it’s about &lt;strong&gt;thinking in harmony with our tools&lt;/strong&gt;. Antigravity IDE embodies that leap: the evolution from editor to intelligent development assistant.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;google&lt;/code&gt;, &lt;code&gt;antigravity&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>google</category>
      <category>antigravity</category>
      <category>programming</category>
    </item>
    <item>
      <title>Antigravity IDE: la nueva generación de editores impulsados por IA</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Tue, 18 Nov 2025 20:13:19 +0000</pubDate>
      <link>https://forem.com/vicodev/antigravity-ide-la-nueva-generacion-de-editores-impulsados-por-ia-56m4</link>
      <guid>https://forem.com/vicodev/antigravity-ide-la-nueva-generacion-de-editores-impulsados-por-ia-56m4</guid>
      <description>&lt;h1&gt;
  
  
  Antigravity IDE: la nueva generación de editores impulsados por IA
&lt;/h1&gt;

&lt;p&gt;La inteligencia artificial ha cambiado la forma en que escribimos, aprendemos y ahora también, cómo programamos. Si VS Code fue el símbolo del desarrollo moderno, &lt;strong&gt;Antigravity IDE&lt;/strong&gt; representa la siguiente frontera: un entorno capaz de comprender al desarrollador, anticipar sus intenciones y colaborar en la creación de software de forma proactiva.&lt;/p&gt;




&lt;h2&gt;
  
  
  De VS Code a Antigravity
&lt;/h2&gt;

&lt;p&gt;El concepto &lt;em&gt;Antigravity&lt;/em&gt; no surge de una herramienta real de Google, sino de una metáfora: la capacidad de elevarse por encima de las limitaciones actuales. En Python, &lt;code&gt;import antigravity&lt;/code&gt; era una broma sobre lo absurdo y genial del código. En este nuevo contexto, Antigravity se convierte en una &lt;strong&gt;visión de desarrollo aumentado por IA&lt;/strong&gt;, donde el editor deja de ser un espacio pasivo para transformarse en un agente inteligente.&lt;/p&gt;

&lt;p&gt;Antigravity IDE combina la flexibilidad de VS Code con la potencia de los modelos de lenguaje, integrando sistemas que aprenden del flujo del usuario y personalizan cada experiencia. Este hipotético fork de Google sería más que una herramienta: &lt;strong&gt;un copiloto autónomo&lt;/strong&gt;, capaz de comprender el contexto, sugerir soluciones y corregir errores en tiempo real.&lt;/p&gt;




&lt;h2&gt;
  
  
  La IA como copiloto consciente
&lt;/h2&gt;

&lt;p&gt;Un IDE con capacidades de aprendizaje profundo podría analizar el comportamiento del programador y adaptar su interfaz y funciones según sus hábitos. Por ejemplo, priorizar comandos o sugerir estructuras de código basadas en proyectos anteriores. &lt;/p&gt;

&lt;p&gt;El sistema sería lo bastante inteligente para entender intenciones, detectar ambigüedades en el código y proponer alternativas sin requerir intervención constante. Este tipo de adaptabilidad rompe el paradigma actual de “autocompletado” y da paso a una &lt;strong&gt;colaboración cognitiva&lt;/strong&gt;, donde la IA aprende del desarrollador tanto como este aprende de ella.&lt;/p&gt;

&lt;p&gt;Imagina un entorno que sabe cuándo estás explorando una idea y cuándo estás optimizando código; que diferencia entre un error lógico y un experimento intencional. Esa es la esencia de Antigravity IDE: un espacio donde la IA &lt;strong&gt;no reemplaza al humano&lt;/strong&gt;, sino que amplifica su creatividad y eficiencia.&lt;/p&gt;




&lt;h2&gt;
  
  
  El valor de un IDE inteligente
&lt;/h2&gt;

&lt;p&gt;La transición de los editores tradicionales a entornos potenciados por IA no es solo una mejora técnica, sino una revolución cultural. Supone un cambio en la relación entre creador y herramienta, donde el editor se convierte en un aliado cognitivo.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Eficiencia sin fricción:&lt;/strong&gt; tareas repetitivas automatizadas, mayor concentración en la lógica.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Aprendizaje continuo:&lt;/strong&gt; el entorno evoluciona según el estilo de cada programador.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Innovación abierta:&lt;/strong&gt; colaboración entre IA, comunidad y código libre.
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;El futuro de la programación no será solo escribir código más rápido, sino &lt;strong&gt;pensar en conjunto con nuestras herramientas&lt;/strong&gt;. Antigravity IDE representa ese salto: el paso de un simple editor a un verdadero asistente de desarrollo inteligente.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;google&lt;/code&gt;, &lt;code&gt;antigravity&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>google</category>
      <category>antigravity</category>
      <category>programming</category>
    </item>
    <item>
      <title>From brute force to reinforcement: optimizing intelligent agents with modern AI</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Tue, 18 Nov 2025 19:19:29 +0000</pubDate>
      <link>https://forem.com/vicodev/from-brute-force-to-reinforcement-optimizing-intelligent-agents-with-modern-ai-4g0k</link>
      <guid>https://forem.com/vicodev/from-brute-force-to-reinforcement-optimizing-intelligent-agents-with-modern-ai-4g0k</guid>
      <description>&lt;h1&gt;
  
  
  From brute force to reinforcement: optimizing intelligent agents with modern AI
&lt;/h1&gt;

&lt;h2&gt;
  
  
  Introduction
&lt;/h2&gt;

&lt;p&gt;Most algorithms start by solving problems with brute force — trying every possible combination until a solution appears. But in real-world scenarios, where the search space can be infinite, this approach becomes infeasible. This is where &lt;strong&gt;reinforcement learning (RL)&lt;/strong&gt; steps in: a branch of artificial intelligence that enables agents to &lt;em&gt;learn from experience&lt;/em&gt;.&lt;/p&gt;




&lt;h2&gt;
  
  
  From brute force to adaptive intelligence
&lt;/h2&gt;

&lt;p&gt;Brute force exhaustively searches for the best solution without learning from failed attempts. RL, on the other hand, is based on &lt;strong&gt;rewards&lt;/strong&gt;: the agent takes actions, observes outcomes, and updates its policy to maximize cumulative reward.&lt;/p&gt;

&lt;h3&gt;
  
  
  Simple example
&lt;/h3&gt;

&lt;p&gt;Imagine a robot trying to find the exit of a maze. Brute force would test every path, while RL learns that “moving north” often leads to a higher reward, refining its policy.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="c1"&gt;# Simplified RL pseudocode
&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;env&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;reset&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;step&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="nf"&gt;range&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;1000&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="n"&gt;action&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;policy&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;new_state&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;reward&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;env&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;step&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;action&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="nf"&gt;update_policy&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;action&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;reward&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;new_state&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;state&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;new_state&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  Core components of reinforcement learning
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Agent:&lt;/strong&gt; the decision-maker.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Environment:&lt;/strong&gt; the world where it acts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reward:&lt;/strong&gt; feedback on performance.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Policy:&lt;/strong&gt; the decision strategy.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;These components let the agent evolve from random behavior to an optimized strategy.&lt;/p&gt;




&lt;h2&gt;
  
  
  Integrating with .NET
&lt;/h2&gt;

&lt;p&gt;The .NET ecosystem allows you to leverage trained models from Python or ONNX. You can use &lt;strong&gt;ML.NET&lt;/strong&gt; or &lt;strong&gt;ONNX Runtime&lt;/strong&gt; to embed RL logic into .NET applications.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime.Tensors&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;

&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;InferenceSession&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"agent.onnx"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;input&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="n"&gt;DenseTensor&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="kt"&gt;float&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="kt"&gt;float&lt;/span&gt;&lt;span class="p"&gt;[]&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="m"&gt;0.1f&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="m"&gt;0.9f&lt;/span&gt; &lt;span class="p"&gt;},&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt;&lt;span class="p"&gt;[]&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="m"&gt;1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="m"&gt;2&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;results&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Run&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="n"&gt;List&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="n"&gt;NamedOnnxValue&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;NamedOnnxValue&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;CreateFromTensor&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"input"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;input&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This approach combines .NET robustness with modern AI flexibility.&lt;/p&gt;




&lt;h2&gt;
  
  
  Real-world applications
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Games:&lt;/strong&gt; agents learning winning strategies.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Route optimization:&lt;/strong&gt; logistics and transport.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finance:&lt;/strong&gt; decision-making based on reward systems.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Final checklist
&lt;/h2&gt;

&lt;p&gt;✅ Understand the difference between brute force and reinforcement.&lt;br&gt;&lt;br&gt;
✅ Build a simple RL environment.&lt;br&gt;&lt;br&gt;
✅ Connect your RL model with a .NET app.&lt;br&gt;&lt;br&gt;
✅ Measure rewards and fine-tune policies.  &lt;/p&gt;




&lt;p&gt;Modern AI isn’t just about solving problems — it’s about &lt;em&gt;learning to solve them better over time&lt;/em&gt;. That’s the essence of reinforcement learning.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;dotnet&lt;/code&gt;, &lt;code&gt;tutorial&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>dotnet</category>
      <category>tutorial</category>
      <category>programming</category>
    </item>
    <item>
      <title>De la fuerza bruta al refuerzo: optimizando agentes inteligentes con IA moderna</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Tue, 18 Nov 2025 19:19:25 +0000</pubDate>
      <link>https://forem.com/vicodev/de-la-fuerza-bruta-al-refuerzo-optimizando-agentes-inteligentes-con-ia-moderna-poa</link>
      <guid>https://forem.com/vicodev/de-la-fuerza-bruta-al-refuerzo-optimizando-agentes-inteligentes-con-ia-moderna-poa</guid>
      <description>&lt;h1&gt;
  
  
  De la fuerza bruta al refuerzo: optimizando agentes inteligentes con IA moderna
&lt;/h1&gt;

&lt;h2&gt;
  
  
  Introducción
&lt;/h2&gt;

&lt;p&gt;La mayoría de los algoritmos comienzan resolviendo problemas con fuerza bruta: probar todas las combinaciones posibles hasta dar con una solución. Pero en el mundo real —donde el espacio de soluciones puede ser infinito— este enfoque se vuelve inviable. Aquí es donde entra en juego el &lt;strong&gt;aprendizaje por refuerzo (Reinforcement Learning, RL)&lt;/strong&gt;, una rama de la inteligencia artificial que permite a los agentes &lt;em&gt;aprender de la experiencia&lt;/em&gt;.&lt;/p&gt;




&lt;h2&gt;
  
  
  De la fuerza bruta a la inteligencia adaptativa
&lt;/h2&gt;

&lt;p&gt;El enfoque de fuerza bruta busca exhaustivamente la mejor solución, sin aprender de los intentos fallidos. En cambio, el RL se basa en &lt;strong&gt;recompensas&lt;/strong&gt;: el agente toma decisiones, observa resultados y ajusta su comportamiento para maximizar su beneficio.&lt;/p&gt;

&lt;h3&gt;
  
  
  Ejemplo básico
&lt;/h3&gt;

&lt;p&gt;Supón que un robot busca la salida de un laberinto. La fuerza bruta probaría cada camino posible, mientras que el RL aprende que “avanzar hacia el norte” suele mejorar la recompensa y ajusta su política.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="c1"&gt;# Pseudocódigo RL simplificado
&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;env&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;reset&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;step&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="nf"&gt;range&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;1000&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="n"&gt;action&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;policy&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;new_state&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;reward&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;env&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;step&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;action&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="nf"&gt;update_policy&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;state&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;action&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;reward&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;new_state&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;state&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;new_state&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  Componentes del aprendizaje por refuerzo
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Agente:&lt;/strong&gt; quien toma decisiones.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Entorno:&lt;/strong&gt; el mundo donde actúa.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Recompensa:&lt;/strong&gt; retroalimentación de su desempeño.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Política:&lt;/strong&gt; estrategia de acción.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Estos elementos permiten que el agente evolucione de un comportamiento aleatorio a uno optimizado.&lt;/p&gt;




&lt;h2&gt;
  
  
  Integración con .NET
&lt;/h2&gt;

&lt;p&gt;El ecosistema .NET ofrece varias formas de aprovechar modelos entrenados en Python u ONNX. Por ejemplo, puedes usar &lt;strong&gt;ML.NET&lt;/strong&gt; o &lt;strong&gt;ONNX Runtime&lt;/strong&gt; para integrar modelos de RL en tus aplicaciones.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime.Tensors&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;

&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;InferenceSession&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"agente.onnx"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;input&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="n"&gt;DenseTensor&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="kt"&gt;float&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="kt"&gt;float&lt;/span&gt;&lt;span class="p"&gt;[]&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="m"&gt;0.1f&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="m"&gt;0.9f&lt;/span&gt; &lt;span class="p"&gt;},&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt;&lt;span class="p"&gt;[]&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="m"&gt;1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="m"&gt;2&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;results&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Run&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="n"&gt;List&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="n"&gt;NamedOnnxValue&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;NamedOnnxValue&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;CreateFromTensor&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"input"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;input&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Con este enfoque, puedes combinar la robustez de .NET con la flexibilidad de los modelos de IA modernos.&lt;/p&gt;




&lt;h2&gt;
  
  
  Casos de uso reales
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Juegos:&lt;/strong&gt; agentes que aprenden estrategias ganadoras.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optimización de rutas:&lt;/strong&gt; logística y transporte.
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finanzas:&lt;/strong&gt; toma de decisiones basada en recompensas.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Checklist final
&lt;/h2&gt;

&lt;p&gt;✅ Comprender la diferencia entre fuerza bruta y refuerzo.&lt;br&gt;&lt;br&gt;
✅ Implementar un entorno RL básico.&lt;br&gt;&lt;br&gt;
✅ Conectar tu modelo de RL con una app .NET.&lt;br&gt;&lt;br&gt;
✅ Medir recompensas y ajustar políticas.  &lt;/p&gt;




&lt;p&gt;La inteligencia artificial moderna no solo busca resolver problemas, sino aprender a resolverlos mejor con el tiempo. Y eso es, precisamente, lo que hace el aprendizaje por refuerzo.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;dotnet&lt;/code&gt;, &lt;code&gt;tutorial&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>dotnet</category>
      <category>tutorial</category>
      <category>programming</category>
    </item>
    <item>
      <title>Integrando IA en aplicaciones .NET: del modelo al despliegue</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Fri, 14 Nov 2025 16:49:10 +0000</pubDate>
      <link>https://forem.com/vicodev/integrando-ia-en-aplicaciones-net-del-modelo-al-despliegue-4ki9</link>
      <guid>https://forem.com/vicodev/integrando-ia-en-aplicaciones-net-del-modelo-al-despliegue-4ki9</guid>
      <description>&lt;h1&gt;
  
  
  Integrando IA en aplicaciones .NET: del modelo al despliegue
&lt;/h1&gt;

&lt;h2&gt;
  
  
  🇪🇸 Español
&lt;/h2&gt;

&lt;p&gt;La inteligencia artificial (IA) dejó de ser una curiosidad académica: hoy es una herramienta práctica que potencia desde chatbots hasta sistemas empresariales complejos. Si trabajas con .NET, la buena noticia es que no necesitas abandonar tu stack para aprovechar su poder. En este artículo aprenderás cómo integrar modelos de IA —entrenados en Python o directamente con ML.NET— dentro de tus aplicaciones .NET, desde el entrenamiento hasta el despliegue.&lt;/p&gt;




&lt;h2&gt;
  
  
  ¿Por qué .NET y no otra plataforma?
&lt;/h2&gt;

&lt;p&gt;.NET ha evolucionado. Con soporte nativo para ML.NET, interoperabilidad con Python y compatibilidad con modelos ONNX, se convierte en un entorno ideal para desarrollar, probar y desplegar soluciones inteligentes sin salir de tu ecosistema habitual.&lt;/p&gt;

&lt;h3&gt;
  
  
  Ventajas principales:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;Integración directa con C# y F#.&lt;/li&gt;
&lt;li&gt;Compatibilidad con modelos ONNX exportados desde frameworks populares.&lt;/li&gt;
&lt;li&gt;Despliegue fácil en entornos Windows, Linux o contenedores Docker.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  1️⃣ Preparando el entorno
&lt;/h2&gt;

&lt;p&gt;Antes de comenzar, asegúrate de tener:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Visual Studio / VS Code&lt;/strong&gt; con SDK de .NET 8+&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ML.NET Model Builder&lt;/strong&gt; o CLI&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Python 3.10+&lt;/strong&gt; si planeas entrenar el modelo fuera de .NET&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ONNX Runtime&lt;/strong&gt; para interoperabilidad&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Ejemplo de instalación rápida:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;dotnet tool &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="nt"&gt;-g&lt;/span&gt; mlnet
pip &lt;span class="nb"&gt;install &lt;/span&gt;onnxruntime scikit-learn
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  2️⃣ Entrenando el modelo
&lt;/h2&gt;

&lt;p&gt;Tienes dos rutas posibles:&lt;/p&gt;

&lt;h3&gt;
  
  
  🔹 Opción A: Usar ML.NET directamente
&lt;/h3&gt;

&lt;p&gt;ML.NET permite crear modelos de clasificación, regresión y detección de anomalías con código C#.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;context&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;MLContext&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;LoadFromTextFile&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="n"&gt;InputData&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;(&lt;/span&gt;&lt;span class="s"&gt;"datos.csv"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;hasHeader&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="k"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;separatorChar&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="sc"&gt;','&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;pipeline&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Transforms&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Conversion&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;MapValueToKey&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"Label"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Transforms&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Text&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;FeaturizeText&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"Features"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s"&gt;"Texto"&lt;/span&gt;&lt;span class="p"&gt;))&lt;/span&gt;
    &lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;MulticlassClassification&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Trainers&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;SdcaMaximumEntropy&lt;/span&gt;&lt;span class="p"&gt;());&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;pipeline&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Fit&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Save&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Schema&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s"&gt;"modelo.zip"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  🔹 Opción B: Entrenar en Python y exportar a ONNX
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;skl2onnx&lt;/span&gt;
&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;sklearn.linear_model&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;LogisticRegression&lt;/span&gt;
&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;skl2onnx&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;convert_sklearn&lt;/span&gt;
&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;skl2onnx.common.data_types&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;FloatTensorType&lt;/span&gt;

&lt;span class="n"&gt;model&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;LogisticRegression&lt;/span&gt;&lt;span class="p"&gt;().&lt;/span&gt;&lt;span class="nf"&gt;fit&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;X_train&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;y_train&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="n"&gt;onnx_model&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;convert_sklearn&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;initial_types&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="p"&gt;[(&lt;/span&gt;&lt;span class="sh"&gt;'&lt;/span&gt;&lt;span class="s"&gt;input&lt;/span&gt;&lt;span class="sh"&gt;'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nc"&gt;FloatTensorType&lt;/span&gt;&lt;span class="p"&gt;([&lt;/span&gt;&lt;span class="bp"&gt;None&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;X_train&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;shape&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;]]))])&lt;/span&gt;
&lt;span class="k"&gt;with&lt;/span&gt; &lt;span class="nf"&gt;open&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;modelo.onnx&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;wb&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;as&lt;/span&gt; &lt;span class="n"&gt;f&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;f&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;write&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;onnx_model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;SerializeToString&lt;/span&gt;&lt;span class="p"&gt;())&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  3️⃣ Integrando el modelo en .NET
&lt;/h2&gt;

&lt;p&gt;Una vez tengas el modelo entrenado (ZIP o ONNX), puedes cargarlo desde C#.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;mlContext&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;MLContext&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;mlContext&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Load&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"modelo.zip"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="k"&gt;out&lt;/span&gt; &lt;span class="n"&gt;_&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;predictor&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;mlContext&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;CreatePredictionEngine&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="n"&gt;InputData&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;OutputData&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;(&lt;/span&gt;&lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;resultado&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;predictor&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Predict&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="n"&gt;InputData&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;Texto&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="s"&gt;"Este es un ejemplo."&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;span class="n"&gt;Console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;WriteLine&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;resultado&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Prediccion&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Para ONNX:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;Microsoft.ML.OnnxRuntime.Tensors&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;InferenceSession&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"modelo.onnx"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="c1"&gt;// Ejecutar inferencia con inputs serializados&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  4️⃣ Desplegando la solución
&lt;/h2&gt;

&lt;p&gt;El modelo puede exponerse como un &lt;strong&gt;servicio web&lt;/strong&gt; con ASP.NET Core:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;app&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;MapPost&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"/predict"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;InputData&lt;/span&gt; &lt;span class="n"&gt;input&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;prediction&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;predictor&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Predict&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;input&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;Results&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Ok&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;prediction&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Y desplegarlo fácilmente en &lt;strong&gt;Docker&lt;/strong&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight docker"&gt;&lt;code&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="s"&gt; mcr.microsoft.com/dotnet/aspnet:8.0&lt;/span&gt;
&lt;span class="k"&gt;COPY&lt;/span&gt;&lt;span class="s"&gt; . /app&lt;/span&gt;
&lt;span class="k"&gt;WORKDIR&lt;/span&gt;&lt;span class="s"&gt; /app&lt;/span&gt;
&lt;span class="k"&gt;ENTRYPOINT&lt;/span&gt;&lt;span class="s"&gt; ["dotnet", "MiApp.dll"]&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  🚀 Buenas prácticas
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Versiona modelos con etiquetas semánticas.&lt;/li&gt;
&lt;li&gt;Registra métricas de rendimiento y deriva de datos.&lt;/li&gt;
&lt;li&gt;Automatiza el retraining mediante pipelines CI/CD.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  🧩 Conclusión
&lt;/h2&gt;

&lt;p&gt;Integrar IA en .NET ya no es un desafío técnico, sino una oportunidad estratégica. Con herramientas maduras y soporte multiplataforma, puedes desarrollar soluciones inteligentes manteniendo tu flujo de trabajo familiar.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;em&gt;“El futuro del desarrollo .NET no solo está en el backend, sino también en la inteligencia.”&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;




&lt;h2&gt;
  
  
  🇬🇧 English
&lt;/h2&gt;

&lt;p&gt;Artificial Intelligence (AI) is no longer an academic curiosity — it’s a practical tool powering everything from chatbots to enterprise systems. If you’re a .NET developer, the great news is: you don’t need to switch tech stacks to take advantage of it. This article walks you through integrating AI models — trained in Python or ML.NET — into .NET apps, from training to deployment.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why .NET?
&lt;/h3&gt;

&lt;p&gt;With built-in ML.NET, Python interop, and ONNX support, .NET provides a friendly environment for intelligent solutions.&lt;/p&gt;

&lt;h3&gt;
  
  
  Steps Covered:
&lt;/h3&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Set up your environment&lt;/strong&gt; (Visual Studio, ML.NET, Python, ONNX Runtime)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Train your model&lt;/strong&gt; (in C# or Python)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Integrate it&lt;/strong&gt; into a .NET app&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deploy&lt;/strong&gt; via ASP.NET Core &amp;amp; Docker&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Machine learning is now native to .NET — bringing AI to your fingertips without leaving your favorite language.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ai&lt;/code&gt;, &lt;code&gt;dotnet&lt;/code&gt;, &lt;code&gt;tutorial&lt;/code&gt;, &lt;code&gt;programming&lt;/code&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>dotnet</category>
      <category>tutorial</category>
      <category>programming</category>
    </item>
    <item>
      <title>From Models to Microservices: Deploying ML in Java and .NET</title>
      <dc:creator>Victor Olvera Thome (Vico)</dc:creator>
      <pubDate>Fri, 14 Nov 2025 15:55:20 +0000</pubDate>
      <link>https://forem.com/vicodev/from-models-to-microservices-deploying-ml-in-java-and-net-en-3nga</link>
      <guid>https://forem.com/vicodev/from-models-to-microservices-deploying-ml-in-java-and-net-en-3nga</guid>
      <description>&lt;h3&gt;
  
  
  Introduction
&lt;/h3&gt;

&lt;p&gt;Deploying Machine Learning models in enterprise environments can be challenging. Moving from a notebook to a scalable service requires more than accuracy: you need integration, maintainability, and automation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why ML Deployment Is Hard
&lt;/h2&gt;

&lt;h3&gt;
  
  
  The gap between research and production
&lt;/h3&gt;

&lt;p&gt;Models work in notebooks but fail on servers. Dependencies, resources, and data evolve.&lt;/p&gt;

&lt;h3&gt;
  
  
  Common pitfalls
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;No versioning for models.&lt;/li&gt;
&lt;li&gt;Manual or unmonitored scaling.&lt;/li&gt;
&lt;li&gt;Lack of automated tests.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Deploying ML with Java
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Using Spring Boot for model serving
&lt;/h3&gt;

&lt;p&gt;Spring Boot allows packaging an ONNX or TensorFlow model as a REST microservice.&lt;/p&gt;

&lt;h3&gt;
  
  
  Integration with ONNX and TensorFlow Java
&lt;/h3&gt;

&lt;p&gt;You can load the model directly and expose it via endpoints.&lt;/p&gt;

&lt;h2&gt;
  
  
  Deploying ML with .NET
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Using ML.NET and API endpoints
&lt;/h3&gt;

&lt;p&gt;ML.NET enables training and serving models directly in C#.&lt;/p&gt;

&lt;h3&gt;
  
  
  Cross-language interoperability
&lt;/h3&gt;

&lt;p&gt;ONNX allows running Python-trained models inside .NET applications.&lt;/p&gt;

&lt;h2&gt;
  
  
  CI/CD for ML Microservices
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Model versioning
&lt;/h3&gt;

&lt;p&gt;Use version control (Git or DVC) to track model and data changes.&lt;/p&gt;

&lt;h3&gt;
  
  
  Automation
&lt;/h3&gt;

&lt;p&gt;Integrate deployment pipelines with GitHub Actions or Azure DevOps.&lt;/p&gt;

&lt;h2&gt;
  
  
  Final Checklist
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;[ ] Is the model versioned and tested?&lt;/li&gt;
&lt;li&gt;[ ] Do the REST endpoints work with real data?&lt;/li&gt;
&lt;li&gt;[ ] Does deployment include monitoring?&lt;/li&gt;
&lt;li&gt;[ ] Is model update automated through CI/CD?&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; &lt;code&gt;ml&lt;/code&gt;, &lt;code&gt;java&lt;/code&gt;, &lt;code&gt;dotnet&lt;/code&gt;, &lt;code&gt;microservices&lt;/code&gt;&lt;/p&gt;

</description>
      <category>machinelearning</category>
      <category>java</category>
      <category>dotnet</category>
      <category>microservices</category>
    </item>
  </channel>
</rss>
