From a Git Clone to an AI-Powered VCS That Understands Your Code

This is a submission for the GitHub Finish-Up-A-Thon Challenge
What I Built
PivotVCS — a full-stack AI-powered version control system built from scratch using Java, Spring Boot, React, and FastAPI. It started as a curiosity project — could I build my own Git-like system? But it grew into something much bigger: a developer tool that not only handles branching, commits, and merges but also understands your code through AI.
PivotVCS lets developers commit files, switch branches, visualize commit history, and now — ask natural language questions about their own codebase. It feels personal because I built every layer of it, from the tree-based file versioning to the RAG pipeline that powers the copilot.
Demo
GitHub Repo Link
BACKEND: https://github.com/JiyaKapoor/Pivot-web
FRONTEND: https://github.com/JiyaKapoor/PivotFrontend
OLD REPO: https://github.com/JiyaKapoor/pivot

Key features in action:
Create branches, commit files, and visualize commit history in a GitHub-style UI
Merge branches with an AI-powered safety analysis before merging
Ask natural language questions about your codebase via the Pivot Copilot panel
Automated code review and branch summarization triggered on every commit

The Comeback Story
When I first built PivotVCS, it was a functional but basic VCS:

-Commit and file versioning
-Branching and merging
-Basic commit history UI

It worked, but it was just another Git clone. Here's what I came back and added to finish it up properly:

1. Database layer for collaboration Added a full persistence layer using PostgreSQL and Hibernate so multiple users can collaborate on the same repository, with optimistic locking on branch HEAD entities to handle concurrent commits safely.
2. Multi-Agent AI Pipeline Integrated a LangGraph multi-agent architecture that automatically triggers on every commit and runs three parallel agents — a code review agent, a branch summarization agent, and a merge safety analysis agent — persisting results per commit for historical querying.
3. RAG-based Copilot Built a full RAG pipeline that indexes repository files as semantic chunks using Mistral AI embeddings, stores them in Supabase pgvector, and powers a natural language chat interface embedded directly in the repo view. Developers can now ask questions like "How is authentication handled?" and get context-aware answers with source file attribution.

My Experience with GitHub Copilot
GitHub Copilot was genuinely useful throughout the finishing phase of this project. The most valuable moments were:

Boilerplate acceleration — setting up DTOs, service layers, and REST controllers in Spring Boot is repetitive; Copilot handled the scaffolding so I could focus on the actual logic

FastAPI and Python interop — switching context between Java and Python frequently, Copilot helped me stay consistent with naming conventions and Pydantic model structures

Debugging assist — When I ran into bugs between the Spring Boot and FastAPI services, Copilot would suggest the fix