MBio AI Chat Docs
Documentation / AI & RAG

RAG Debugger

See exactly which chunks were retrieved for a question and diagnose weak or irrelevant knowledge.

Test a real question

Enter a question a visitor might ask. The debugger shows total retrieval score, vector similarity, keyword score, source title and URL.

Tune retrieval

Use RAG Settings to adjust Top K, minimum similarity, candidate limits, vector/keyword weights, chunk size, overlap and maximum context size. Retest the same query after each meaningful change.

Knowledge gaps

If the best retrieval score stays below the configured gap threshold, the live chatbot can record that visitor question in Knowledge Gaps for review.