A secure, production-ready AI app that transforms YouTube videos into interactive learning experiences with quizzes, code exercises, and Google Colab integration.
- Visual Video Analysis: Upload any coding tutorial and Gemini AI analyzes both audio and visual content to understand what's being taught 🧠
- Auto-Generated Coding Exercises: Creates interactive coding challenges based on the tutorial content that you can solve in real-time 💻
- Built-in Code Compiler: Test and run generated exercises directly in your browser - no external setup needed 🔥
- AI Chat Assistant: Ask questions about the video content and get instant answers from the AI that watched your tutorial ✨
- Custom Learning Materials: Generate personalized flashcards, quizzes, and coding exercises tailored to the video content 📓
- Interactive Learning Hub: Everything in one place - video player, code editor, AI chat, and learning materials 🎨
- Real-time Execution: Run code, get feedback, and iterate instantly while learning ⚡
| Tag | Description |
|---|---|
beta |
Public Beta |
# Pull the image
docker pull imhari14/haash-lm:beta
# Run container
docker run -p 3000:3000 -p 8001:8001 imhari14/haash-lm:betadocker run -d -p 3000:3000 -p 8001:8001 --name haashlm imhari14/haash-lm:beta
- Frontend: http://localhost:3000
- Health Check: http://localhost:8001/health
- Start container
- Open http://localhost:3000
- Configure API keys in settings
Required Keys:
- Gemini API Key (https://aistudio.google.com/app/apikey)
- Judge0 API Key (https://judge0.com/ or https://rapidapi.com/judge0-official/api/judge0-ce)
| Feature | HaashLM | AI Studio | Improvement |
|---|---|---|---|
| Max Video Length | 90 minutes | 40 minutes | +125% |
| Token Consumption | ~150K tokens | 700K-800K tokens | 5x Less |
| Video Processing Method | Proprietary Smart Sampling | 1FPS extraction | 10x Efficient |
| Initial Processing Time | 3 minutes (90min video) | 3 minutes (40min video) | 3x Faster |
| Consecutive Responses | 30-40 seconds | 3+ minutes | 5x Faster |
| Context Handling | Maintains low hallucination (fewer tokens) | Hallucinates >500K | Rock Solid |
Why HaashLM handles longer videos: While AI Studio can technically upload 1-hour videos, it fills up the context window 100%, leaving no space for chat replies - making 40 minutes the practical limit for interactive use.
🚀 Here's the game-changer: HaashLM processes a full 90-minute video using only 150k tokens - that's incredible efficiency! Our advanced video processing algorithms intelligently sample key frames and optimize content analysis, allowing you to learn from feature-length tutorials while maintaining plenty of context space for meaningful conversations.
- Video Processing: Transcription, summarization, multi-language support
- AI-Powered Generation: Quizzes, flashcards, code exercises, Colab notebooks
- Code Execution: Multi-language support, real-time execution, Judge0 integration
- Modern Interface: Responsive design, dark/light themes, intuitive UX
- Frontend: Next.js 14 + TypeScript + Tailwind CSS
- Backend: FastAPI + Python 3.12
- AI Integration: Google Gemini 2.5 Flash API
- Code Execution: Judge0 CE API
- Video Processing: FFmpeg + yt-dlp
- Security: Obfuscated bytecode, non-root user
# Health check
curl http://localhost:8001/health
# Logs
docker logs -f haashlm- Invalid API key: Check format, permissions, and quota
- Port conflicts: Ensure 3000 & 8001 are free
- System Config: Minimum 8-12 core CPUs and 8GB RAM for smooth experience
- Network issues: Check connectivity to APIs
We’d love to hear your thoughts and suggestions to improve HaashLM!
Please take a minute to fill out our short feedback form:
⭐ Star this repo if you find it useful!