
Every hotel front desk fields the same calls all day: more towels, a toothbrush, what time is checkout, the Uber Eats rider is downstairs. Today a staff member answers, writes the request down, collects the items and walks them up to the room, leaving the desk empty while guests queue in the lobby. Many calls come in a language the person on shift doesn't speak, and even hotels with delivery robots still need a human to take the call and type in the room number. Concierge makes the phone call the whole workflow. A guest dials the front desk and just talks, in English, Mandarin or a mix. The AssemblyAI voice agent already knows the room from the switchboard and gets to work: it orders amenities and food against the live menu and stock, offering a substitute when something runs out; books a delivery-app order waiting at the desk onto the same robot trip; answers hotel FAQs with exact figures, from checkout times and late-checkout fees to gym, pool and restaurant hours, Wi-Fi, breakfast, parking and laundry; tracks an order by room with a live ETA, changes it mid-flight or recalls the robot; and escalates what needs a human, like a late checkout, a lost key card, a broken AC or a complaint, to the front-desk queue, then hangs up when the guest says goodbye. Every request becomes a real mission for a delivery robot simulated in MuJoCo: it turns to the counter while staff load the bin, drives the corridor, rings the room's phone to bring the guest to the door, opens its bin, then heads home and parks. AssemblyAI runs the entire conversation in one real-time session: speech recognition, turn detection tuned for noisy rooms, keyterm prompting for room numbers and dish names, barge-in, and client-side tools that act on live robot state, with our own LLM via OpenRouter. A live ops dashboard shows the call as it happens, the fleet, every delivery, the call log, escalations and inventory, and the web demo runs voice and physics in the browser.
30 Sep 2026