GitOdyssey

Inspiration

In large organizations, onboarding new developers to massive codebases can take months. Even experienced engineers spend significant time tracing commits, PRs, and diffs just to understand how a feature evolved. We realized this is a huge productivity bottleneck and in the world of AI, where 6 months can mean an entirely new generation of technology, that delay is unacceptable.
So, we built GitOdyssey a tool to help developers learn their codebase at the speed of thought.

What it does

GitOdyssey transforms a Git repository into an interactive, AI-powered knowledge graph. It uses semantic search and commit-level summarization to let developers:

Ask natural-language questions about their codebase (e.g., “When was authentication refactored?”).
Explore diffs, PRs, and commits visually through a connected commit graph.
Understand why code changed not just what changed.

How we built it

Backend: Python + FastAPI + Gemini 2.5 Flash
Indexing: We combined keyword search and vector embeddings via Google Gemini for hybrid retrieval.
Frontend: React + Tailwind + Vite for a clean and fast developer experience, featuring an interactive commit visualization and semantic query interface.
Infrastructure: Supabase for hosting our Postgres database and Renderer for hosting our backend server and frontend

Challenges we ran into

Traversing and storing large commit histories efficiently without bloating storage.
Designing a schema that supports both semantic and keyword search.
Balancing retrieval accuracy and performance in our hybrid index.
Building an intuitive UI to visualize complex Git histories.

Accomplishments that we're proud of

Built a fully functional prototype in under 36 hours.
Designed a scalable ingestion pipeline that can walk entire repositories.
Integrated natural-language search to make exploring commits as easy as asking a question.
Created a beautiful and fast React interface for visual commit exploration.