← All work
Case StudyTier BPrototypeSpring 2023

Bilingual Semantic Medicine Search

Persian/English semantic retrieval prototype built from Iranian FDA and WHO ATC data.

AIDataResearch

Context

Spring 2023 · Academic prototype · Maturity: Prototype.

Problem

Users needed cross-language semantic retrieval over medicine data rather than exact keyword matching.

Users

Academic prototype context.

My Role

Academic NLP product/implementation.

Product Decisions & Responsibilities

  • Framed the system as information retrieval, not medical advice.

Technical Approach

  • Built a 65,469-record combined dataset and bilingual semantic retrieval flow.
NLPSemantic searchIranian FDA dataWHO ATC data

Constraints

  • No comprehensive trustworthy F1/accuracy metric in current evidence.
  • Healthcare domain requires careful scope.

Product Flow

  1. Prepare FDA/WHO data
  2. Normalize bilingual fields
  3. Encode/index
  4. Semantic query
  5. Return ranked records

Outcome

  • Final dataset contained 65,469 records.

What I Learned

  • In high-stakes domains, defining what the product does not do is part of product design.

Evidence Boundary

What this does — and does not — prove

Retrieval prototype only — not diagnosis, prescription, or treatment recommendation.

Next Iteration

Extend the product only after the next decision can be tied to stronger user, operational, or business evidence rather than adding features for their own sake.