Car Dealership RAG

A sales-expert agent that answers car-buying questions and narrows your options. It is not stubborn with its suggestions, and it uses conversation history to keep your context across turns. The agent is built in Rust with rig and pgvector. Retrieval pairs structured SQL range filters with semantic vector search over embeddings, ranked by cosine similarity. Ranges enforce precise constraints like price, mileage, and year. Vectors capture the fuzzier "what kind of car" intent. Together they return a tight, grounded slice of the inventory for the LLM to reason over.

View source →