Production RAG (advanced)
Hybrid retrieval, reranker, response caching, eval feedback and guardrails.
Rendering…
Make it your own.
flowchart LR
Q[User query] --> G1[Guardrail:\nharm / PII / topic]
G1 --> RW[Query rewrite]
RW --> Hybrid[Hybrid retrieval]
KW[(BM25 / lexical)] --> Hybrid
Vec[(Vector store)] --> Hybrid
Hybrid --> RR[Reranker]
RR --> Ctx[Top-K context]
Cache[(Response cache)] -.->|hit| Reply
RW -.->|cache key| Cache
Ctx --> P[Prompt template]
Q --> P
P --> LLM[LLM]
LLM --> G2[Guardrail:\nfaithfulness + citations]
G2 --> Reply[Answer + citations]
Reply --> Tel[Telemetry]
Tel --> Eval[Eval harness]