Skip to content
← All cases
Ship the AI

Git-RAG-Chatbot – Local RAG Git Assistant for Developers

A local, Docker-based RAG (Retrieval-Augmented Generation) chatbot enabling developers to query internal Git repositories in natural language. The system periodically indexes repositories, generates embeddings with LLM models, and stores them in ChromaDB as a vector store with rich code context metadata.

client
Personal Project / Adaptive Biotechnologies
when
2024
headline
Natural-language search over private repos
  1. 01 · Problem

    Searching large codebases for specific implementations, understanding legacy code, and generating patches is time-consuming. RAG enables developers to ask natural-language questions over code, understand context across files, and generate/refine code using enterprise LLMs—all running locally for security and privacy.

  2. 02 · What I built

    • Built repository indexing system that chunks source files and generates embeddings with LLM embedding model
    • Enriched chunks with metadata (repo, file path, language, symbol/function context, commit info)
    • Implemented vector search over ChromaDB for context-aware code retrieval
    • Developed Gradio chat UI backed by Python service for natural-language code queries
    • Integrated with Codex CLI and ChatGPT Enterprise for code generation and analysis
    • Containerized with Docker for local deployment, privacy, and developer productivity
  3. 03 · Result

    • Natural-language code search across Git repositories
    • Faster onboarding and codebase understanding
    • Privacy-first local deployment with Docker
    • Production use for internal developer productivity

Related work