Case study · Search infrastructure
Search that matches on meaning, not keywords.
Clinicians rarely type the exact title of what they need. I integrated AI semantic search across a medical publisher's whole content library, and it became the retrieval layer that keeps their clinical AI assistant grounded.
The problem
A doctor typing "kidney patient on a gliptin" means the renal-dosing section of the sitagliptin monograph. Keyword search sees no matching title and returns nothing, or returns pages that share a word but not the intent. Every miss erodes trust, and for an AI assistant a bad retrieval guarantees a bad answer.
Try it
Pick a query to see where it lands in the embedding space, and what keyword search would have found instead.
What I built
- One semantic index spanning drug monographs, disease information, clinical news, special and conference reports, and education courses.
- Embedding-based retrieval so queries match on intent: synonyms, drug classes and loosely described patients all resolve to the right content.
- Cross-type ranking that puts the most relevant item first, whether it's a monograph section, a news story or a course.
- The retrieval layer the clinical AI assistant relies on to fetch grounded context and cite it.
Why it matters
The quality of an AI answer is capped by the quality of what you retrieve.
Semantic search turned the content library into something both people and the AI assistant could reliably navigate. It's the foundation that let the assistant promise "every answer draws only on our own data" and keep that promise.
Outcome
Loosely phrased questions still land on the right content, and the same index powers both direct search and grounded, cited AI answers.