I'm 12 and Built an AI Platform with RAG, Memory & Tools in 52 Hours — Plus a Full Ecosystem (Browser, Mail, Calendar, Chat, Search)

Arnav Gupta — Sun, 08 Mar 2026 17:05:25 +0000

By Arnav Gupta
April 2026

So here's the thing — I keep seeing posts about "I built X in Y hours" and honestly? Most of them are cap.

So let me be real with you.

I built Wizard.AI — a full AI platform with RAG, persistent memory, web search, code execution, image generation, 7 personalities, user accounts, cloud sync, custom community personalities, developer hub, multi-agent studio, turbo mode, and more.

Then I kept building.

Now there's also:

🧙 Wizard Browser — Electron-based with workspaces, tab groups, ad blocking, password manager, and cloud sync
📧 Wizard Mail — Full email client with alias support (@wizardecosystem.dpdns.org)
📅 Wizard Calendar — Event management with Gmail sync
💬 Wizard Chat — Real-time messaging between Wizard users
🔍 Wizard Search — AI-powered search page at wizardecosystem.dpdns.org/search
And coming soon:
🐧 ArcWizOS — A Linux distribution built from scratch, designed for users who value privacy and AI workflows (targeting late 2026)

Not one marathon session. Not 24 hours straight. Just consistent building when I had time.

And no, the first version wasn't some masterpiece. It was a terminal bot that took me 1 hour to throw together.

Wait, What Does It Actually Do?
Let me break down what Wizard.AI v13.0 looks like today:

🧠 7 AI Personalities Switch between JARVIS, ORACLE, Nerd, Fun, Sarcastic, Fast, and Normal modes. Each one has its own system prompt and personality. Talk to JARVIS like Tony Stark's AI, get mystical with ORACLE, or keep it quick with Fast mode.
🎭 Custom Community Personalities Users can create and share their own AI personalities. Want a pirate-themed AI? A math tutor? A coding assistant? Make it, share it, and let others use it. Each personality tracks likes and uses. **
📚 RAG (Retrieval-Augmented Generation)** Upload PDFs, text files, documents — anything. Wizard.AI chunks them, extracts keywords, and stores them. Ask questions about your documents and get answers with context. It's like having a personal knowledge base.
🧠 Persistent Memory Here's the thing most chatbots don't do — I actually remember you. Tell me your name once, and I'll remember it forever. Tell me your age, hobbies, preferences — stored in a database and recalled across every conversation. No more repeating yourself.
🌐 Web Search Ask about current events, and I'll search the web in real-time using Tavily. The AI automatically decides when to search based on your query — no manual toggle needed.
💻 Code Execution Write Python code, run it in a sandbox, see the output. Test snippets, learn to code, or just mess around. It's sandboxed so you can't break anything.
🎨 Image Generation Describe what you want, and I'll generate it using Stable Diffusion. Choose from square, landscape, or portrait sizes.
🤖 Multi-Agent Studio Chain specialized AI agents together — Researcher, Writer, Coder, Designer, Data Analyst, Reviewer, Summarizer, Translator, Optimizer. Run pre-built workflows or build your own custom chains.

🧙 Wizard Browser

Built on Electron with everything you'd expect:
Tab groups with drag-and-drop reordering
Workspaces (Default, Work, School, Media)
Brave-like settings (ad blocking, private mode, password manager)
AI-powered search with real summaries
Cloud sync across devices
Bookmark management

📧 Wizard Mail

Full email client that connects to any IMAP provider (Gmail, Outlook, Yahoo):
Alias support — Get @wizardecosystem.dpdns.org email addresses that forward to your real inbox
Rich text composer with AI enhancement
Image upload and file attachments
Reply threading with proper headers
Folder management (Inbox, Sent, Starred, Spam, Trash)

📅 Wizard Calendar

Schedule management with clean month/week views:
Create, edit, delete events
Color-coded categories
Mini calendar with event indicators

💬 Wizard Chat

Real-time messaging between Wizard users:
Search for users by email
Instant message delivery
Typing indicators
Read receipts
Persistent conversation history

🔍 Wizard Search

Google-like search page at wizardecosystem.dpdns.org/search:
AI-powered summaries
Search suggestions
Direct result navigation (no redirects)
DuckDuckGo + Wizard.AI backend

- 📊 Stats Dashboard
Track your usage — messages sent, files uploaded, images generated, searches performed, memories stored. See your average response time and daily activity.

- ⚡ Turbo Mode
Flip the switch and get responses up to 2x faster. Uses optimized models and higher token limits.

- 🔐 User Accounts
Sign up with email verification, log in, and your chats sync across devices. Everything saves to the cloud. Guest mode works too.

- 🎤 Voice Input
Click the microphone and speak your message. Works in all modes.

- 💬 Multi-Chat
Create multiple conversations, rename them, delete them, switch between them. Each chat keeps its own history and personality.

- 📱 PWA + Mobile Ready
Install it on your phone like a native app. Works offline, loads instantly, feels native.

The Ecosystem at a Glance

            Service URL                    What it does

Wizard AI wizardai.dpdns.org Main AI platform
Wizard Browser (Electron app) AI-powered browser
Wizard Mail wizardecosystem.dpdns.org/mail Email client
Wizard Calendar wizardecosystem.dpdns.org/calendar Event management
Wizard Chat wizardecosystem.dpdns.org/chat Messaging
Wizard Search wizardecosystem.dpdns.org/search AI search page
ArcWizOS (Coming 2026) Linux distro
The Tech Stack
Frontend:

HTML, CSS, JavaScript (vanilla, no frameworks)
Custom purple theme with glass morphism
Electron for desktop browser
PWA with service worker for offline support
Custom domain: wizardai.dpdns.org / wizardecosystem.dpdns.org

Backend:

Python + Flask
SQLAlchemy + SQLite
Groq API (Llama 3.1 8B and 3.3 70B)
Custom RAG with keyword search
Tavily API for web search
Hugging Face for image generation
IMAP/SMTP for email (no Google OAuth required)
Cloudflare API for email aliases

Key Features:

Session auth with 30-day persistence
Memory extraction using regex
File upload with duplicate detection
Stats tracking per user
Streaming responses (token-by-token)
Auto web search detection
Community personalities browser
Real-time chat with polling

The Code That Makes It Work
Here's the streaming endpoint that ties everything together:

python

@app.route('/api/chat/stream', methods=['POST'])
def stream_chat():
    user_id = session.get('user_id')
    data = request.json

    def generate():
        full_prompt = build_prompt(user_id, data['mode'], data['prompt'])

        client = Groq(api_key=app.config['GROQ_API_KEY'])
        stream = client.chat.completions.create(
            model="llama-3.1-8b-instant",
            messages=[{"role": "user", "content": full_prompt}],
            stream=True
        )

        for chunk in stream:
            if chunk.choices[0].delta.content:
                yield f"data: {json.dumps({'token': chunk.choices[0].delta.content})}\n\n"

    return Response(stream_with_context(generate()), mimetype='text/event-stream')

And the custom personality system:

python

class Personality(db.Model):
    id = db.Column(db.Integer, primary_key=True)
    user_id = db.Column(db.Integer, db.ForeignKey('user.id'))
    name = db.Column(db.String(50), nullable=False)
    emoji = db.Column(db.String(10), default='🤖')
    system_prompt = db.Column(db.Text, nullable=False)
    is_public = db.Column(db.Boolean, default=False)
    likes = db.Column(db.Integer, default=0)
    uses = db.Column(db.Integer, default=0)

What I Learned

Start ugly
My first version was a terminal script. Embarrassing to show anyone. But it worked, and that's what matters. You can't polish something that doesn't exist.
Databases aren't scary
I was terrified of SQLAlchemy at first. Then I just started typing and Googling errors. Now it's fine. Most things are like this.
Memory changes everything
The moment the AI remembered my name across conversations, it stopped feeling like a chatbot and started feeling like an assistant. That one feature gets more positive feedback than anything else.
50 hours is enough to build something real
You don't need months. You don't need a team. Just consistent time and actually finishing things.
Custom personalities unlock community
The feature I spent the least time on (custom personalities) got the most excitement. People love creating and sharing their own AIs.
An ecosystem > a single app
Adding email, calendar, chat, and search created a complete workspace. Users stay within the ecosystem instead of jumping between Google services.
Age doesn't matter
Nobody cares that I'm 12 when they're using the product. They just care if it works. The internet is a meritocracy — build something useful and people will use it.

Try It Yourself
👉 Wizard AI: https://www.wizardai.dpdns.org
👉 Wizard Ecosystem: https://wizardecosystem.dpdns.org
👉 Github: https://github.com/AG-Ultima

Desktop App: Download from github — workspaces, tab groups, ad blocking, password manager, cloud sync.

If you find bugs (you will), have ideas, or just want to say hi — drop a comment or reach out on Twitter.

What's Next
I'm keeping going. v13.0 is live, but v14.0 is already in my head:

🐧 ArcWizOS — A Linux distribution built from scratch. Think Arch Linux meets AI workstation. Custom kernel, Wayland compositor, and deep Wizard.AI integration. Target release: late 2026. Tryna make ChromeOS cry here.
Enterprise features — Outlook/Teams integration, SSO, audit logs
Mobile app — React Native for iOS/Android
Plugin system — So users can add custom tools
Fine-tuned models — For each personality

Built by Arnav, age 12, in about 52 hours spread over a month (plus the ecosystem in another 20 hours).

Probably should have done homework. Worth it. 🧙

Forem: Arnav Gupta

I'm 12 and Built an AI Platform with RAG, Memory & Tools in 52 Hours — Plus a Full Ecosystem (Browser, Mail, Calendar, Chat, Search)