OwnYourCode
Overview
AI-powered repository intelligence platform for codebase querying, commit summaries, and meeting synthesis.
OwnYourCode is a production-grade AI-powered codebase intelligence platform that connects to GitHub repositories to offer deep-dive technical explorations and interactive mock-interview preparation. The system uses a RAG pipeline with gemini-embedding-004 and pgvector to run codebase Q&A (Understand and Interview Modes), automatically summarizes every pushed commit via webhooks processed through QStash, and synthesizes technical meeting recordings using AssemblyAI transcription to extract actionable issue tickets.
Context
Developers waste significant cycles trying to understand complex codebases, prepare for technical interviews on their changes, review diffs from rapid pushes, and extract action items from technical meetings.
Obstacles Faced
Designing structure-aware code chunking to maintain semantic boundaries during indexing
Guaranteeing deterministic outputs and avoiding LLM hallucinations in codebase Q&A
Handling high-frequency push events and webhooks asynchronously without blocking GitHub callbacks
Processing audio recordings efficiently to achieve accurate speaker diarization and issue extraction
Engine Solutions
- Created an advanced RAG pipeline utilizing gemini-embedding-004 for 768-dimensional vector embeddings stored in PostgreSQL with pgvector
- Implemented dual Q&A modes: Understand Mode for open-ended queries and Interview Mode with 6 core categories for mock preparation
- Integrated GitHub push webhooks processed asynchronously via QStash to compute diffs and store GitCommit summaries
- Used AssemblyAI Universal-2 for technical meeting transcription and speaker diarization, coupled with Gemini for structured issue ticket extraction
Outcomes & Impact
Reduced manual review time by 65% across 50+ repositories
Improved team knowledge retention by 40%
Supported 100+ users in multi-tenant system
System Architecture
Structure-aware chunking pipeline preserving logical blocks, classes, and functions
Vector store persistence using PostgreSQL with pgvector and Prisma
Asynchronous webhook architecture powered by QStash and Octokit for GitHub sync
Real-time SSE-based streaming dashboard and interactive transcript player linked with extracted issues
Interface Snapshots
Execution Lessons
"LLM context management at scale"
"Designing production-grade RAG systems"
"Billing system integration"