Project overview
A Gemini-backed product assistant for a rice brand. Answers catalogue questions, suggests recipes and helps customers choose between products.
A product assistant for a premium rice brand, built to answer the questions a customer would otherwise ask a shop assistant: which variety suits what they are cooking, how one product differs from another, and what to actually make with it. It runs on the Gemini API behind a Node and Express backend.
The part that matters is that it does not answer from whatever the model happens to know about rice. A knowledge base of plain markdown files describing the real product range and a set of recipes is loaded and keyword-indexed at startup, and the matching passages are what the model is prompted with, so updating what the assistant knows means editing a document rather than retraining anything. It ships as an embeddable widget for an existing storefront, and a Twilio webhook lets the same assistant answer over WhatsApp.