CodeVec is a hackathon project that brings intelligent code understanding to life using natural language queries.
It functions like a real-time CodeLLM assistant: you can ask questions about your codebase, and it responds based on the actual code context—no need for manual prompt engineering. Just write code, ask, and get answers.
- Understands your code with context-aware LLMs
- Query using natural language
- Supports full projects
- Powered by Ollama-hosted models
- Real-time backend interaction
1st Runner-Up at the hackathon where it was built.
Built to explore smarter code interaction through LLMs.
| Service | Core Language / Tech | Purpose |
|---|---|---|
| Studio | TypeScript (Next.js) | Frontend UI for chatting and interacting with the code assistant. |
| Backend | Python (FastAPI) | Orchestrates requests, talks to Redis, Supabase, and triggers Celery jobs. |
| Supabase | PgSQL | Stores metadata, user sessions, and project information. |
| Redis | Redis | Acts as a task queue for Celery and also streams LLM output to client. |
| Celery | Python (Celery) | Executes background tasks like LLM inference. |
| Stream Proxy | Python (FastAPI) | Forwards the stream response from redis to client with proper authentication |