2025
CodeRevU — AI-Powered GitHub PR Review
An AI-powered GitHub pull request review platform that automatically generates structured, actionable code reviews using Retrieval-Augmented Generation (RAG) and Gemini AI.
Technology Stack
Next.js 16TypeScriptPostgreSQLPrismaGemini AIPineconeInngestGitHub Webhooks
Overview
System Architecture
- •Frontend: Next.js 16 App Router for a high-performance, server-rendered dashboard.
- •Backend & Orchestration: Inngest powers reliable, event-driven workflows (e.g., 'pr.created') that run asynchronously to handle long-running indexing tasks.
- •Data & Vector Store: PostgreSQL (via Prisma) stores user/project data, while Pinecone indexes code embeddings for semantic retrieval.
- •AI Engine: Google Gemini Pro generates code reviews based on retrieved contexts, grounded by static analysis data.
Key Challenges
- •Indexing large repositories without blocking GitHub webhook flows.
- •Designing a RAG pipeline that retrieves semantically relevant code instead of naive file matches.
- •Handling concurrent PR events safely across multiple repositories.
- •Preventing hallucinated feedback by grounding LLM responses in real code context.
Key Learnings
- •Pinecone-based vector search significantly improved contextual relevance over keyword-based approaches.
- •Asynchronous workflows with Inngest prevented API blocking and enabled safe concurrency.
- •Structured prompts reduced noisy or vague review output.
- •Webhook-driven architectures require idempotency and replay safety.
Uniqueness
- •PR-level RAG instead of generic repo chat.
- •Fully automated review generation on PR open/update.
- •Structured feedback designed for real engineering teams.
Impact
- •Eliminated repetitive manual review effort for common PR patterns.
- •Enabled faster review cycles with consistent feedback quality.