AR AI Museum Guide

A conversational museum guide on a phone. The Unity AR app recognises the exhibit in front of the camera, and the visitor asks about it out loud. Answers come back as speech, grounded in that exhibit and fact-checked with Google Search grounding.

There are two voice modes: a quick record-and-answer pipeline over REST, and a live mode that streams audio both ways over WebSockets to Gemini, so visitors can interrupt and ask follow-ups.

API keys never touch the phone: a Node.js and PostgreSQL backend on Railway makes every AI call, and spoken answers are cached per exhibit so repeat questions don't spend model time. I gathered requirements in client meetings, deployed through GitHub Actions and gave live demos to the client.

Commissioned by Khora during my placement, January to May 2026. I was the lead and only developer.