Retail catalogues
Visual Similarity Search
Find visually similar products from a photo, using image embeddings and vector search.
The problem
Shoppers often know what a product looks like but not what it is called, so the catalogue needed to be searchable by image.
What I built
- A FastAPI service using CLIP image embeddings and Qdrant vector search.
- Celery workers that download and index images asynchronously, with S3 storage.
- A React front end, Prometheus metrics and structured logs.
- An earlier FAISS-based prototype for a furnishings catalogue.