video1.0<iframe src="https://www.loom.com/embed/c63c3a5668e2440e9dc67829299714f4" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/c63c3a5668e2440e9dc67829299714f4-433c53aa5a235dbe.gif293.604This Loom explains how to build and test a patient-style voice bot for the AI Engineering Challenge using a local Python FastAPI WebSocket setup. The system opens a WebSocket, immediately sends a hello message to trigger the clinic bot, then runs a loop that receives Retail AI responses, stores response IDs and transcripts, and forwards them to Cloud to generate replies, using Cloud Haiku by Entropiq. It also discusses why WebSockets are used for ultra-low latency and the author’s approach to testing with multiple personas: Confusing Rambling Patient, Anxious and Panicked One, Distracted Parent, and Aggressive Adult Child. For testing, it requires starting the JVCorn server and Ngrok, running a dialer script to trigger a call and download call artifacts, including the transcript and an OGG audio file after the call ends.