Hybrid RAG System
A production-grade retrieval-augmented generation pipeline with hybrid retrieval, cross-encoder reranking, context compression, and RAGAS evaluation.
Documents
…
Chunks
…
Eval runs
…
📄 Ingest
Upload PDF/Markdown/text. Context-aware chunking preserves document structure.
Open →
💬 Ask
Cited answers with the full retrieval debug: BM25, vector, fusion, rerank.
Open →
📊 Evaluate
RAGAS scorecards and vector-only vs hybrid+rerank comparisons.
Open →
Documents → Loader (PDF/MD/TXT) → Chunker → BGE Embeddings → ChromaDB (vector) + BM25 Index (keyword) Query → Vector Search + BM25 → RRF Fusion → Reranker → Compression → Prompt + Citations → LLM → Answer