DevScribe

Inspiration

The biggest hurdle when joining a new software project is spending days or weeks just trying to understand a large, unfamiliar codebase. We were inspired to build a tool that acts as an instant expert, helping developers get up to speed in minutes, not weeks.

What it does

DevScribe is an AI assistant that lets you "chat" with any public GitHub repository. You provide a URL, and it allows you to ask complex questions about the code in plain English—like "How does user authentication work?"—and get answers with context directly from the source code.

How we built it

We built a full-stack application using a React frontend and a Python backend. The backend uses FastAPI for the web server and LangChain to orchestrate the AI logic. We used Google's Gemini API for embeddings and chat, with FAISS as a local vector database to store and retrieve code context.

Challenges we ran into

We faced several real-world bugs. Our biggest challenge was a conflict between our server's auto-reloader and the file system, which caused the server to restart during processing. We also had to debug numerous API issues with git and the Google Gemini API, from incorrect model names to quota limits, and had to re-architect our data loading to ensure the AI saw all the code files.

Accomplishments that we're proud of

We're proud of building a fully functional, end-to-end RAG (Retrieval-Augmented Generation) pipeline from scratch. The app successfully ingests entire codebases and provides genuinely helpful, context-aware answers. We're also proud of the polished UI, especially the syntax-highlighted code blocks in the chat, which makes it feel like a professional tool.

What we learned

We learned a huge amount about the practical side of building AI applications. This included the importance of robust error handling between a client and server (sending proper HTTP status codes), the subtleties of different LangChain modules, and how to systematically debug and iterate on API and model availability issues with LLM providers.

What's next for DevScribe

We'd love to add support for private GitHub repositories via OAuth authentication. Other future steps include moving to a persistent vector database to save analyzed repos for later use, and integrating DevScribe as a VS Code extension to bring the AI assistant directly into a developer's editor.