Agentic AI operating system to auto-pilot culinary workflows.
Every hospitality SaaS claims to solve inventory, yet line cooks routinely ignore them. In commercial kitchens, grease coats every surface, the ambient noise hits 110dB during dinner service, and chefs have about 3 seconds to log trimmings before their pans catch fire. Traditional enterprise software built with 5-step dropdown modal wizards simply dies here.
“I’ve got 14 orders on the pass and hot clarified butter boiling over. If your tablet makes me choose between 'Spanish Onion (Rough)' and 'Spanish Onion (Fine)' from a dropdown, that tablet is getting tossed in the compactor.”
We spent two weeks on the line at commercial prep commissaries in Sydney. We watched prep cooks take off vinyl gloves just to fight a capacitive touchscreen, mistype weights, and eventually give up. That failure cost restaurant groups over $220,000 in unlogged protein and produce spoilage every year.
Instead of asking cooks to type, we turned the tablet into a quiet companion appliance. We tethered industrial digital scales via Bluetooth Low Energy (BLE) running on dedicated Web Workers. When a cook drops a prep container onto the scale, the weight locks automatically in 120ms. The tablet contextually detects the station and suggests the 3 most likely trimmings with massive 84px hit targets.
Continuous byte stream parsing weight offsets and stable tare signals without blocking the main UI thread.
Building great software is mostly about deciding what *not* to build. Here is what we originally thought would work, why it collapsed in real usability testing, and what we replaced it with.
Earlier in the project, leadership wanted an AI vision camera that would 'automatically classify vegetables on the cutting board using vision models'. It sounded futuristic in a slide deck, but in reality it was catastrophic AI slop.
We threw away the camera entirely. We built a rock-solid, zero-inference physical integration: Bluetooth tare scale + giant 84px station preset pads + immediate thermal ticket printing. No LLM hallucinations, zero latency.
Try the real component states below. Everything runs client-side in React with real state machines.