Product Similarity API

Semantic search with real constraints

Pure vector search returns things that are similar. Shoppers want things that are similar and under a price, and above a rating. That combination — semantic recall with structured filtering — is where most naive implementations fall over, returning either irrelevant matches or an empty set.

Embeddings are stored in Chroma with metadata alongside, so filters apply against the candidate set rather than after it. Served over FastAPI.