jev-perfume-advisor

Scent Sommelier is a perfume-recommendation chatbot that uses TypeSafe Jev, a model that returns typed, calibrated decisions instead of generated text, over fragrance data in FragDB format. The application is available as a GitHub repository and can be run locally via Node.js or Docker. It accepts natural-language requests such as "Suggest me a perfume for an evening party" or "Something for Nordic winters" and provides recommendations along with a "How Jev decided" panel that shows each typed decision Jev made, with its probability. The system is designed so that code owns the control flow and Jev supplies semantic judgements, following TypeSafe's recommended approach. Jev is described as a System One decision model, not a chat model, and it never generates text. The bot's replies are constructed from deterministic templates that turn typed decisions into prose, ensuring that every perfume fact comes from the catalog and cannot be invented.