The RAG Full Application Production Blueprint V3 defines the architecture for a next-generation retrieval-augmented generation pipeline.
It incorporates advanced techniques such as Hybrid Search, which combines keyword-based BM25 and semantic pgvector search.
Another key technique is ColBERT token-level scoring (MaxSim) which allows for deep contextual evaluations.
The frontend uses Vite and React with a Green/Violet aesthetic theme.
Agentic RAG utilizes a smart agent (Qwen3) that uses tools like search_docs and get_page to dynamically find answers.