Lauren — a customer-services avatar
A real-time avatar customer-services agent served to a 55-inch touchscreen kiosk and run locally on a single GPU. Lauren talks, answers, and helps front-of-house with a real voice and an animated face — and hands off to a human when she should. No per-word cloud cost, low latency, all on the box in the room.
The idea
Front-of-house AI that people actually talk to. A customer walks up to the kiosk and speaks or taps; Lauren answers as a consistent, branded face and voice. Everything runs on the local kiosk GPU, so responses are fast, running cost is flat, and no conversation has to leave the building.
How it works — the real-time render pipeline
Lauren reuses the platform's own real-time avatar stack: a low-latency call loop, a self-hosted voice, a warm always-ready brain, and a GPU-accelerated animated face, wired into one render loop on the kiosk.
Built in
Real-time avatar
A GPU-accelerated animated face (real-time, TensorRT) driven by the platform's self-hosted, punctuation-timed voice, with a warm always-ready brain for instant replies. Stylised fallback if photoreal will not hold frame-rate on the card.
Safeguarding alert system
Wellbeing or safeguarding signals pause the avatar and route to a person — never handled autonomously. Every interaction is auditable.
Daily whitelisted knowledge refresh
An automated daily lookup to a whitelisted business-information source keeps Lauren current on services, hours, events and offers — refreshed once a day into her knowledge base, so there is no per-query cloud call and she always knows what is on.
Human handoff
Lauren knows the edge of her competence and passes real people to real people. Bookings and enquiries are drafted and confirmed, never committed blind.
Delivery
Lauren installs from the platform the same way the back-office agents do — a public base plus the tenant's private bolt-ons — so any business could stand up its own branded front-of-house avatar.