<?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: Siddharth</title>
    <description>The latest articles on Forem by Siddharth (@s_b_yrrah).</description>
    <link>https://forem.com/s_b_yrrah</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%2F3665856%2Fa6d73db6-a83e-4a73-90a4-c677ff00c860.jpg</url>
      <title>Forem: Siddharth</title>
      <link>https://forem.com/s_b_yrrah</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://forem.com/feed/s_b_yrrah"/>
    <language>en</language>
    <item>
      <title>Most tools feel heavier than they need to be</title>
      <dc:creator>Siddharth</dc:creator>
      <pubDate>Mon, 22 Dec 2025 10:53:18 +0000</pubDate>
      <link>https://forem.com/s_b_yrrah/most-tools-feel-heavier-than-they-need-to-be-5ep2</link>
      <guid>https://forem.com/s_b_yrrah/most-tools-feel-heavier-than-they-need-to-be-5ep2</guid>
      <description>&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%2Fi91tn8c7kjqhkmoqxpv3.jpeg" 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%2Fi91tn8c7kjqhkmoqxpv3.jpeg" alt=" " width="800" height="338"&gt;&lt;/a&gt;&lt;br&gt;
Most software doesn’t feel heavy because it’s slow or badly engineered.&lt;br&gt;
It feels heavy because it asks for attention too often.&lt;/p&gt;

&lt;p&gt;Each new feature usually makes sense in isolation. But over time, these decisions stack up — more UI states to understand, more options to consider, more interruptions to flow.&lt;/p&gt;

&lt;p&gt;I’ve been thinking a lot about how lightness in tools isn’t about removing power.&lt;br&gt;
It’s about reducing unnecessary decisions.&lt;/p&gt;

&lt;p&gt;Curious how others notice this in their daily tools.&lt;/p&gt;

</description>
      <category>ux</category>
      <category>discuss</category>
      <category>design</category>
      <category>productivity</category>
    </item>
    <item>
      <title>From a Pyodide Wrapper to an Edge AI Engine</title>
      <dc:creator>Siddharth</dc:creator>
      <pubDate>Wed, 17 Dec 2025 01:00:51 +0000</pubDate>
      <link>https://forem.com/s_b_yrrah/from-a-pyodide-wrapper-to-an-edge-ai-engine-1ejl</link>
      <guid>https://forem.com/s_b_yrrah/from-a-pyodide-wrapper-to-an-edge-ai-engine-1ejl</guid>
      <description>&lt;p&gt;A few months ago, I shipped &lt;strong&gt;v1&lt;/strong&gt; of a React library that let you run &lt;strong&gt;Python ML models directly in the browser&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;It worked — but the feedback was blunt and fair:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;em&gt;“This is basically just a wrapper around Pyodide.”&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That line mattered, because it exposed the real problem.&lt;/p&gt;




&lt;h3&gt;
  
  
  &lt;em&gt;What v1 got right (and wrong)&lt;/em&gt;
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;v1 proved feasibility&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Python + WASM + React &lt;em&gt;can&lt;/em&gt; work client-side.&lt;/p&gt;

&lt;p&gt;But it also broke down fast in real apps:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;Inference froze the UI&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;Workers outlived React components → memory leaks&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;WASM OOMs caused silent hangs&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;Cleanup logic was fragile and scattered&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Running Python wasn’t the hard part.&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Orchestrating failure was.&lt;/strong&gt;&lt;/p&gt;




&lt;h3&gt;
  
  
  &lt;em&gt;What changed in v2&lt;/em&gt;
&lt;/h3&gt;

&lt;p&gt;v2 is a full rewrite — shifting from a &lt;em&gt;wrapper&lt;/em&gt; to an &lt;em&gt;infrastructure layer&lt;/em&gt; for Edge ML:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Supervised worker pools&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Hung or OOM workers are terminated and replaced automatically.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Strict React lifecycle binding&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Component unmount = worker termination. No zombies.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Zero-copy data paths&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
SharedArrayBuffer where possible to avoid serializing large tensors.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The goal wasn’t speed alone — it was &lt;em&gt;predictability under failure&lt;/em&gt;.&lt;/p&gt;




&lt;h3&gt;
  
  
  &lt;em&gt;Where this is going&lt;/em&gt;
&lt;/h3&gt;

&lt;p&gt;I’m now prototyping a &lt;strong&gt;Neural Bundler&lt;/strong&gt; — a build-time compiler that translates Python math logic into &lt;strong&gt;WebGPU compute shaders&lt;/strong&gt;, skipping Pyodide entirely for math-heavy workloads.&lt;/p&gt;

&lt;p&gt;Still early. Very promising.&lt;/p&gt;




&lt;p&gt;If you’re building &lt;strong&gt;client-side ML&lt;/strong&gt;, edge inference, or anything involving &lt;strong&gt;WASM + React&lt;/strong&gt;, I’d love your thoughts.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/ShyamSathish005/python-react-ml" rel="noopener noreferrer"&gt;Repo link: https://github.com/ShyamSathish005/python-react-ml&lt;/a&gt;&lt;/p&gt;

</description>
      <category>react</category>
      <category>python</category>
      <category>architecture</category>
      <category>machinelearning</category>
    </item>
  </channel>
</rss>
