KnowledgeLoop

KnowledgeLoop

RAG you can measure — not just demo.

Citation-backed answers over your docs, with confidence gating and a reproducible evaluation harness.

Try the live demo 7 seed docs indexed

Ask with citations

Retrieve from your Markdown and PDF docs, answer with sources, refuse when unsure.

Public demo

No account required — chat against the Northstar Runbooks seed corpus.

Measurable quality

Hit-rate, latency, and refusal metrics on the case study.

How it works

  1. Ingest — Upload Markdown or PDF; chunk and embed with OpenAI.
  2. Retrieve — Search an on-disk Chroma index for the top-k passages.
  3. Answer — Generate with citations; low confidence becomes an explicit refusal.
  4. Evaluate — Re-run a fixed golden set and publish metrics.