Case Study

MediSwap

MediSwap finds composition-based medicine alternatives with fuzzy matching, semantic search, FastAPI, Qdrant, and embeddings.

Open Live Demo View Code
MediSwap interface preview

Problem

The challenge

Medicine search by name alone can miss useful composition-equivalent alternatives when spelling, naming, or phrasing differs.

Solution

What I built

I built a semantic recommendation system that combines fuzzy matching with vector similarity over embeddings to surface composition-based medicine alternatives.

Architecture

How it works

  • RapidFuzz handles fast fuzzy lookup across medicine names and related text.
  • BGE embeddings and Qdrant support semantic similarity for composition-aware retrieval.
  • FastAPI exposes the recommendation flow for a deployable Hugging Face Spaces app.

Implementation

Project highlights

  • Built a medicine recommendation platform for composition-based alternatives.
  • Combined fuzzy string matching with vector similarity over BGE embeddings.
  • Architected the retrieval layer with FastAPI, SQLite, RapidFuzz, and Qdrant.