SciencesLoop Open to AI Engineer roles
Healthcare AI · reliability research

Guideline-faithfulness chat

A guideline recommendation carries an official strengthstrong, conditional — set by the guideline's expert committee. Ask a knee-osteoarthritis question and this demo reports what NICE NG226 and ACR/AF say, re-reading each source to keep that strength faithful — instead of flattening two guidelines into a single confident-sounding answer. Retrieve → answer → re-read source & correct.

Research demo of guideline faithfulness — not medical advice, not a clinical tool. It reports published-guideline positions with citations; it does not recommend treatment.

Part of GuidelineBench-OA, a benchmark measuring whether AI faithfully restates clinical-guideline recommendations. This public demo covers NICE + ACR/AF for knee osteoarthritis.