Hey Product Hunt! 👋 I'm Çağrı Giray Keşan, a Full-Stack & AI Systems Engineer and author of Hybrid RAG Issue & PR Assistant. While developing code intelligence and repository tools, I noticed a critical challenge in codebase RAG: dense embeddings alone often struggle with exact token matches like variable names, error codes, and configuration flags because identifiers get blurred in high-dimensional latent space. To address this, I built **Hybrid RAG**: - ⚡ **Lexical Search:** Embedded SQLite FTS5 (Okapi BM25) for sub-millisecond exact token precision with zero cloud database dependency. - 🧠 **Semantic Search:** Dense vector embeddings for contextual semantic intent. - 🔀 **Rank Fusion:** Reciprocal Rank Fusion (k=60) to blend lexical and dense scores deterministically. - 🤖 **AST Code Triage:** Automatic repository issue and PR triage with markdown AST-aware chunking and verified line-span citations ([file#L-L]). The core recipe is merged in microsoft/PhiCookBook (PR #571) and is packaged as a zero-dependency GitHub Marketplace Action via @vercel/ncc. - 📦 GitHub Marketplace: https://github.com/marketplace/a... - 📖 Engineering Deep-Dive: https://dev.to/cagrik34/i-ditche... Looking forward to your thoughts and feedback! 🚀

Hybrid RAG

Zero-cloud SQLite FTS5 RAG engine & GitHub Action

Zero-cloud SQLite FTS5 RAG engine & GitHub Action