🧠 AxiomAI - Intelligent LLM Router

Smart prompt routing system that optimizes costs by automatically selecting between different LLM tiers based on complexity.

🌟 Overview

AxiomAI is an intelligent routing system that analyzes prompt complexity and automatically routes requests to the most cost-effective LLM tier. By using ML classification and rule-based heuristics, it can reduce LLM costs by up to 70% while maintaining response quality.

✨ Key Features

• 🎯 Smart Routing: Hybrid ML + heuristics-based classification (98% accuracy)
• 💰 Cost Optimization: Automatic tier selection saves up to 70% on API costs
• 🔐 Secure Authentication: Supabase-powered auth with email, OAuth (Google/GitHub), and OTP
• 📊 Analytics Dashboard: Real-time usage statistics and routing insights
• ⚡ Fast API: Deployed on HuggingFace Spaces with optimized inference
• 🌐 Chrome Extension: Route prompts directly from any AI chat interface

🚀 Live Deployment

• Frontend: https://axiomai1918.vercel.app
• Backend API: https://ritesh1918-axiom-backend.hf.space
• API Docs: https://ritesh1918-axiom-backend.hf.space/docs

🏗️ Architecture

┌─────────────┐      ┌──────────────┐      ┌─────────────┐
│   Frontend  │─────▶│   Backend    │─────▶│   Supabase  │
│   (Vercel)  │      │ (HuggingFace)│      │  (Database) │
└─────────────┘      └──────────────┘      └─────────────┘
                            │
                            ▼
                     ┌──────────────┐
                     │  ML Classifier│
                     │ (HF Inference)│
                     └──────────────┘

🛠️ Tech Stack

Frontend

• Framework: Vanilla HTML, CSS, JavaScript
• Authentication: Supabase Auth (Email, OAuth, OTP)
• Deployment: Vercel
• UI: Modern glassmorphism design with responsive layout

Backend

• Framework: FastAPI (Python)
• ML Model: Fine-tuned DistilBERT (98% accuracy)
• Database: PostgreSQL (Supabase)
• Deployment: HuggingFace Spaces
• API: RESTful with auto-generated OpenAPI docs

ML Model

• Base Model: DistilBERT-base-uncased
• Training Data: 1,000 balanced samples
• Accuracy: 98%
• Inference: HuggingFace Inference API

📦 Project Structure

AxiomAI/
├── frontend/                 # Frontend application
│   ├── index.html            # Landing page
│   ├── login.html            # Authentication page
│   ├── dashboard.html        # Main dashboard
│   ├── css/
│   │   └── style.css         # Styles
│   ├── js/
│   │   ├── auth.js           # Authentication logic
│   │   ├── dashboard.js      # Dashboard functionality
│   │   ├── charts.js         # Analytics charts
│   │   └── supabase.js       # Supabase client
│   └── assets/               # Static assets
│
├── extension/                # Chrome extension
│   ├── manifest.json         # Extension config
│   ├── popup.js              # Popup logic
│   └── content.js            # Content script
│
└── docs/                     # Documentation
    ├── SETUP.md              # Setup guide
    └── SUPABASE_SETUP.md     # Database configuration

🚀 Quick Start

Prerequisites

• Node.js (optional, for development server)
• Supabase account
• Modern web browser

Frontend Deployment (Vercel)

1. Fork/Clone this repository

   git clone https://github.com/ritesh-1918/AxiomAI.git
   cd AxiomAI

2. Deploy to Vercel

   • Connect your GitHub repo to Vercel
   • Set root directory to /
   • Deploy automatically

3. Configure Supabase (see docs/SUPABASE_SETUP.md)

   • Create Supabase project
   • Run database schema
   • Update frontend/js/supabase.js with your credentials

Local Development

# Navigate to frontend
cd frontend

# Start local server (Python)
python -m http.server 8001

# Or use any static server
# npx serve .

Visit: http://localhost:8001

🔧 Configuration

Supabase Credentials

Update frontend/js/supabase.js:

const SUPABASE_URL = 'your-project-url';
const SUPABASE_ANON_KEY = 'your-anon-key';

Backend API URL

Update frontend/js/dashboard.js:

const API_URL = 'https://ritesh1918-axiom-backend.hf.space';

📊 Features

Smart Routing

• ML Classification: Fine-tuned DistilBERT model
• Heuristics: Rule-based patterns for code, length, keywords
• Confidence Scoring: Transparent routing decisions

Authentication

• Email/Password: Standard authentication
• OAuth Providers: Google and GitHub
• Magic Links: Passwordless OTP login
• Row-Level Security: Secure data access

Analytics Dashboard

• Real-time Stats: Requests, tier distribution, latency
• Charts: Visual routing insights
• System Logs: Live activity monitoring
• Cost Tracking: Estimated savings

Chrome Extension

• One-Click Routing: Route prompts from any AI interface
• Visual Feedback: See routing decisions instantly
• Seamless Integration: Works with ChatGPT, Claude, etc.

🌐 API Endpoints

Backend API Documentation

Visit: https://ritesh1918-axiom-backend.hf.space/docs

Main Endpoints:

• POST /api/v1/route - Route a prompt to appropriate LLM tier
• GET /api/v1/stats - Get user routing statistics (requires auth)
• GET /health - Health check

Example Request:

curl -X POST https://ritesh1918-axiom-backend.hf.space/api/v1/route \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Write a Python function"}'

Response:

{
  "selected_tier": "LARGE_LLM",
  "confidence": 0.95,
  "latency_ms": 87,
  "routing_reason": "Code generation detected"
}

🎯 Use Cases

1. Individual Developers: Reduce personal LLM API costs
2. Startups: Optimize infrastructure spending
3. Enterprises: Smart routing for large-scale deployments
4. Research: Study prompt complexity patterns
5. Education: Learn about ML classification systems

📈 Performance

• Routing Accuracy: 98%
• Average Latency: < 100ms
• Cost Savings: Up to 70%
• Uptime: 99.9% (HuggingFace Spaces)

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🔗 Links

• Live Demo: axiomai1918.vercel.app
• Backend API: ritesh1918-axiom-backend.hf.space
• Documentation: docs/SETUP.md
• GitHub: github.com/ritesh-1918/AxiomAI

👨‍💻 Author

Ritesh

• GitHub: @ritesh-1918
• Email: ritesh1918@users.noreply.github.com

🙏 Acknowledgments

• HuggingFace for model hosting and inference
• Supabase for authentication and database
• Vercel for frontend hosting
• DistilBERT team for the base model

---

Made with ❤️ for optimizing LLM costs