Tote: Hands-Free Voice Picking for Warehouses

In a warehouse, pickers walk the aisles collecting items for customer orders. It's the most expensive job on the floor, and most of it still runs on paper lists or handheld scanners that keep a hand busy and pull the picker's eyes off the shelf. Voice systems exist, but workers have to train them on their voice first and learn a fixed list of commands, usually in one language. Tote is the voice in the picker's headset. There's nothing to train and no commands to learn: Tote says where to go: "Aisle A, bay 3, level 2." At the shelf, the picker reads out the two check digits printed on the label. If they're at the wrong shelf, Tote catches it before anything is picked. The picker just talks. "Only three here" asks for a restock, "this one's damaged" puts it on hold, and "there's a spill in aisle F" closes the aisle and warns the supervisor. An urgent order can re-route the picker while they walk. It works in English, Spanish, German, French, Italian and Portuguese. The shift lead gets a voice assistant too, Tote Desk. They can ask "How's Sam doing?", say "the spill is cleaned up" to reopen the aisle, or say "tell him great pace" and Tote passes it on in the picker's language. How Tote uses AssemblyAI: Voice Agent API: runs both live conversations, the picker's and the lead's. Streaming speech recognition: tuned for product names, warehouse words and noisy floors. Sessions API: keeps a recording of every shift. Universal-3 Pro: listens to the recording again after the shift and double-checks every check-digit read. The AI only decides what the picker meant. Our code reads the actual numbers from what was said, so a check digit can't be made up. We tested full shifts against the live API in all six languages and on a noisy floor: 64 of 64 picks right. Tote replies about 1.6 seconds after the picker stops talking, down from 4 seconds. Voice costs about 3 cents per pick.