Docs2DB-API - RAG Server For a Docs2DB Database

Docs2DB-API is an RAG server for use with Docs2DB databases.

See https://github.com/rhel-lightspeed/docs2db

See demos/llama-stack/README.md

Summary:

• Gather a folder of your source data documents
• Run Docs2DB on that folder to get a database
• Import Docs2DB-API in your LLM applications to use your source data in RAG.

RAG Algorithm

Docs2DB-API implements modern retrieval techniques:

• Contextual chunks with document-level context prepended to each chunk
• Hybrid search combining BM25 (lexical) and vector embeddings (semantic)
• Reciprocal Rank Fusion (RRF) for result combination
• Cross-encoder reranker for improved result quality
• PostgreSQL full-text search with tsvector and GIN indexing
• pgvector for similarity search
• Granite embedding models (30M parameters, 384 dimensions)
• Normalized model metadata storage
• Question refinement for improved query expansion
• Universal RAG engine adaptable to multiple API frameworks