LumiBot Watches the Hands, Not the Robots
A camera over the workbench. A model trained on a single motion. LumiBot catches the mistake at the moment it's made - not at the end of the line where it already cost you.
On most factory floors, the moment of truth arrives late. A worker assembles a unit, it slides down the line, and only at final quality control does someone notice the wrong screw, the missed clip, the part seated backward. By then the mistake has traveled - and travel is expensive. LumiBot, a Y Combinator Winter 2022 company, is built on a simple relocation of that moment: catch the error where it happens, at the hands.
The current product is an AI copilot for manual assembly. A camera sits over the workbench and watches the operator's hands. Computer vision reads the motion, compares it against the correct sequence, and flags a deviation in real time - before the unit moves on. The company's own line for it is plain: "Ship faster than ever, error-free."
01 / THE IDEAInspection at the source
There's an old rule in manufacturing that the cost of a defect climbs the further it travels from where it was made. Catch it at the station and you re-seat a part. Catch it at final QC and you tear down a finished unit. Catch it after shipping and you have a recall. LumiBot's bet is that the cheapest place to inspect is the place everyone finds hardest to watch: the exact second a human hand does the work.
That framing matters because the easy story about factory AI is replacement - robots taking the bench. LumiBot's approach is the opposite. The operator stays. The AI watches, learns the workflow, and speaks up only when something looks off. It's a copilot, not a replacement, which also happens to make it cheaper to adopt: no new production line, no robot arms, just a camera and a model tuned to your process.
02 / HOW IT WORKSConsult, demo, deploy
LumiBot doesn't sell a shrink-wrapped box. Onboarding runs in three phases, and each one lowers the risk of the next.
The models are bespoke. Rather than a generic "assembly detector," LumiBot fine-tunes on the specific motions of a specific workflow - which is what lets a single camera tell a correct step from a near-miss. The more work it sees, the sharper it gets.
03 / THE PIVOTFrom spreadsheets to hands
LumiBot didn't start here. In its YC W22 form, the company described itself as "AI-powered bots for industrial backoffice automation" - a human-in-the-loop service that chewed through repetitive data ops, operations and revenue-ops support, content moderation, AI training and verification, and custom process automation. There was a procurement and MRO-inventory flavor too: supplier management, multi-vendor parts ordering, price monitoring, no-code workflows.
The through-line from then to now is productivity, not a particular technology. The company moved from automating the paperwork around manufacturing to watching the manufacturing itself. Same north star, different medium - spreadsheets gave way to a camera pointed at a workbench.
04 / THE FOUNDERA second act in industrial AI
LumiBot is led by founder and CEO Aswin Andrison. Before this, he consulted at McKinsey, IBM and Honeywell, and co-founded STOQO - a fresh-produce supply startup that, at its height, served roughly 40,000 restaurants with a workforce around 1,500 and about USD 13M raised. He studied computer science and chemical engineering at UNSW, then added an MBA across London Business School and MIT.
That resume reads like the setup for exactly this product: enterprise consulting taught him where industrial pain lives, and running a high-volume physical-goods operation taught him what a single error costs when it multiplies across a line.
05 / THE BACKERSWho's betting on it
Beyond Y Combinator, LumiBot's early-stage backing includes Monk's Hill Ventures, Alpha JWC Ventures, Goodwater Capital, SWC Global and January Capital - a lineup that reads distinctly Southeast Asia, matching the company's Singapore base and regional roots. The thesis they're buying is human-plus-AI industrial productivity, delivered without asking a factory to rebuild anything.
06 / THE MARKETCopilots for the physical world
Manufacturing computer vision isn't empty - names like Instrumental, Elementary, Invisible AI and Retrocausal work nearby corners of quality and assembly monitoring. LumiBot's distinguishing move is where it aims the lens: not the finished part on a conveyor, but the operator's hands mid-task, with models fine-tuned to a specific motion. It sits in the broader shift of AI copilots leaving the software world and showing up on the shop floor, over someone's shoulder, quiet until it isn't.
For a manufacturer, the pitch reduces to something testable: point one camera at your hardest manual step, let LumiBot learn it, and see whether the defect that used to surface at final QC surfaces at the bench instead. If it does, the math takes care of itself.