Robodyx

Architecture & Process

Context

Trainees log numbers diligently and then have no idea what to change, and food logging is the part everyone abandons first because typing out a meal is tedious. Robodyx puts an AI on both ends: one that reads a meal into a log entry, and one the trainee can actually talk to about it. Operators get a board across the whole roster.

Role

End to end: the React app shipped to devices through Capacitor, the Express API on GCP, the Firebase data layer, both AI integrations, and the admin portal the operators run it from.

Key Engineering Decision

Two different models for two different jobs. Food logging goes through Vertex AI as a one-shot extraction — a meal in, a structured log entry out — while the coach runs on Gemini Live so the trainee can hold a real conversation. Forcing both through one model would have made the logging chatty and the coach stilted.