How answers are retrieved
Setting upCareWise searches an approved corpus of clinical documents by meaning and keywords, selects the most relevant passages, and uses them to ground its answer — so claims trace back to a real source.
No passages indexed yet — run `npm run ingest` to enable evidence-backed answers.
How retrieval works
Each approved document is split into passages and stored with semantic embeddings (using Voyage AI). When a parent asks a question, CareWise retrieves the most semantically similar approved passages and uses them to construct a grounded answer.
Retrieval is restricted to the approved corpus. CareWise cannot browse the web or retrieve from unapproved sources.
When retrieval finds nothing
If no relevant approved passages are found, CareWise uses a fallback: "I don't have a trusted source for this question." It never invents an answer when evidence is missing.