rohanps630/ai-code-reviewer

AI Code Reviewer is a portfolio project: an AI agent that analyzes GitHub pull requests to surface issues, references, and test outcomes using code-aware retrieval and tool-using agent loops. Intended users are engineers, repo maintainers, and AI-integration engineers who want an audit-capable, production-oriented PR reviewer. Key features: - Code-aware retrieval: AST-based chunking with tree-sitter, hybrid BM25 + pgvector search, cross-encoder reranking. - Agentic loop: ReAct-style tool agent that searches code, reads files, finds references, and runs tests. - Evals: golden dataset of 50+ historical PRs, LLM-as-judge plus deterministic checks. - Production concerns: prompt & semantic caching, model routing, prompt-injection defenses, Langfuse/Sentry tracing and cost optimizations. Stack and layout: Next.js frontend, Python indexer, Postgres + pgvector, Drizzle ORM, Anthropic (primary) + OpenAI fallback, Voyage embeddings, Cohere rerank. Repo organized into apps/web, apps/indexer, packages/agent, packages/db, packages/shared, and scripts/cli.mjs.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.