gitRAG

RAG-based GitHub Repo Analysis Platform
Analyse any public GitHub repository with LLM-powered chat and advanced semantic search.

---

Untitled.video.-.Made.with.Clipchamp.2.mp4

⭐ Overview

Situation

As a participant in open-source competitions and project exhibitions (EPICS, university projects), I often struggled to deeply understand large codebases—especially when onboarding new repositories from group members or exploring unfamiliar open-source projects. Sifting through thousands of files, dependencies, and scattered documentation was tedious and overwhelming, making it hard to answer even basic questions like "Where is X implemented?" or "How does this module work?"

Task

I needed a platform that would let me:

• Instantly chat with any GitHub repo to ask questions about code, architecture, or logic.
• Quickly visualize and explore repo structure, file contents, and metadata.
• Perform semantic code search (not just by filename/text).
• Support multiple users and projects securely for my team and in competitions.

Action

I independently designed and built gitRAG—an end-to-end, multi-tenant platform that ingests any public GitHub repo, chunks and indexes its code using embeddings and vector search, and enables users to interactively chat, search, and analyse codebases using a modern LLM (via LangChain and OpenAI API).

• Built secure, scalable backend using FastAPI, PostgreSQL (Aiven), PineconeDB, and LangChain.
• Developed a modern React frontend with hierarchical file explorer, real-time AI chat, and repo analytics.
• Integrated Google/GitHub OAuth2 for authentication, and per-user encrypted API key management for privacy.
• Engineered ingestion pipelines to chunk, embed, and index 50MB+ codebases with 10,000+ files.
• Tested and deployed the platform on multiple real-world repos for open-source events and university project groups.

Result

• Significantly reduced onboarding time for new repositories—now get context, explanations, and code Q&A in seconds.
• Enabled my team and myself to confidently tackle larger, more complex projects in hackathons and coursework.
• gitRAG is now a robust, reusable tool for anyone needing rapid understanding of unfamiliar codebases.

---

🚀 Features

• LLM-powered code chat: Ask questions about repo structure, functions, or files—get contextual, AI-driven answers.
• Semantic code search: Find relevant code snippets using meaning, not just keywords.
• Hierarchical file explorer: Browse and preview the full repo tree with metadata and analytics.
• Multi-user & multi-repo support: Secure, per-user data isolation with Google/GitHub OAuth2.
• Repo analytics: Visualize language breakdown, file types, contributors, and more.
• Encrypted API key management: User API keys are encrypted and never exposed.
• Blazing fast: Sub-second query responses (vector search and retrieval).
• Modern UI: Built with React, TailwindCSS, and Three.js (for 3D hero effect).

---

🛠️ Tech Stack

• Frontend: React.js, TailwindCSS, Vite, Three.js
• Backend: FastAPI (Python), LangChain, PostgreSQL (Aiven), PineconeDB
• AI/Vector Search: OpenAI API, PineconeDB, LangChain
• Auth: Google OAuth2, GitHub OAuth2
• Integrations: GitHub API (repo fetching, metadata), Node.js (utility scripts)

---

📷 Demo

---

⚡ How it Works (RAG Pipeline)

1. Login with Google or GitHub OAuth2 (secure, per-user).
2. Paste any public GitHub repo URL and your OpenAI API key (encrypted).
3. Ingestion:
   • Fetches repo files via GitHub API
   • Chunks code using custom logic (by file type/size)
   • Generates vector embeddings (LangChain + OpenAI API)
   • Stores chunks and metadata in PineconeDB and PostgreSQL
4. Analysis & Chat:
   • Use AI chat to ask any question about the repo (“What does X function do?” “Show me auth logic”)
   • Semantic search finds and retrieves the most relevant code chunks
   • LLM (via LangChain) generates contextual, accurate answers using retrieved code
5. Explore:
   • Hierarchical explorer shows real file tree, lets you preview content and metadata
   • Repo analytics panel for high-level insights

---

🧩 Architecture

✨ Example Use Cases

• Hackathons/open-source events: Instantly understand any team repo or competition project.
• University coursework: Quickly onboard and analyze group project submissions.
• Personal learning: Explore popular open-source projects by chatting and searching their code.
• Team code reviews: Get instant explanations and context for PRs and legacy code.