Quorev is an LLM-powered product recommendation engine for e-commerce. This page is the detailed version of what we're building, how, and what's next.
Most e-commerce recommendation systems are collaborative-filtering or rules-based: "customers who bought X also bought Y." They work on volume but can't handle a real sentence like "something warm for a rainy camping trip, not too bulky" — and when they get a recommendation wrong, nobody can say why. Lean commerce teams without an ML org are stuck choosing between generic bundled recommenders or building this themselves.
Three stages, each handled by a distinct, inspectable component rather than one opaque model call:
An LLM converts a free-text query, order history, or support transcript into structured constraints: category, budget, occasion, must-have/must-avoid attributes.
Embeddings-based retrieval (RAG) over the merchant's actual product catalog and inventory — so results are things that exist and can ship, not plausible-sounding hallucinations.
Every ranked result carries the constraints that produced it, so both the shopper and the merchant's team can see and audit why it was picked.