Answer
Enablement Assistant RAG
Ask the notebook. Inspect the evidence.
A static product demo of the runnable RAG prototype. It keeps the same contract as the CLI: retrieve trusted markdown chunks, answer only from the indexed sources, cite line ranges, and refuse unsupported questions.
Markdown corpus
Chunk metadata
Lexical retrieval
Grounded answer
Retrieval
Retrieved Context
Eval Set
Regression Questions
18/18
current evals pass in workspace validation
11
covered questions expect citations
7
uncovered question expects refusal
Production Path
What Changes Next
- Swap lexical retrieval for hybrid embeddings and keyword search.
- Add document freshness, permissions, and stable source URLs.
- Use a model gateway with grounded-answer and refusal evals.
- Publish latency, cost, source-hit, and answer-quality telemetry.