CocoChat — Side project

Most database tools assume you already know the query. CocoChat started from the opposite end: describe what you want, let the assistant find the shape of it, and keep the whole thing conversational.

Next.js on the front, MongoDB behind it, Firebase handling auth so sessions were solved rather than invented. The transcript renders incrementally, which matters more to how fast it feels than any actual latency number.

The personality is a thin layer, on purpose. It sits in the copy and the loading states, never in the answers.

In a chat interface, perceived speed is a rendering problem before it is a backend problem. Streaming the transcript did more than any query optimisation.