Niobia takes the guesswork out of some of tech’s most expensive mistakes

San Francisco-based startup Niobia is betting large connected a captious but often overlooked portion of artillery and spot making: helping engineers recognize however and wherefore things spell wrong. Its AI bundle flags accumulation defects and lets engineers inquire a chatbot what caused them, aiming to present answers successful minutes alternatively than days.

How it works broadly: The company’s customers spot AI agents connected their accumulation lines to look for defects, past loop successful quality engineers to enactment retired what mightiness origin a merchandise to fail. The agents tally connected self-contained GPUs mounted straight connected manufacturing equipment, overmuch similar a sensor. They tin spot pinhole-sized flaws specified arsenic scratches. Niobia’s AI besides gathers each the information successful 1 spot truthful engineers tin inquire a chatbot what went incorrect and excavation deeper into the cause.

Speed is truthful captious due to the fact that chips and batteries are immoderate of the world’s astir important products, portion besides progressively successful abbreviated supply. As manufacturers proceed to make and amended their technologies, things inevitably spell awry during probe and development. Engineers past walk days oregon longer figuring retired a hole earlier they tin restart production. Each flaw oregon mistake tin beryllium costly — sometimes moving into the millions of dollars — and acceptable backmost manufacturing timelines.

Niobia co-founders Gaurav Jha and Harveen Kaur some person PhDs successful chemistry and person worked astatine chipmakers and artillery manufacturers, giving them firsthand acquisition with the challenges wrong the industry. In a telephone up of TechCrunch Disrupt 2026, wherever the institution is simply a Battlefield finalist, Jha explained that portion what Niobia does is specialized, it serves a multibillion-dollar marketplace that includes immoderate of the biggest artillery and semiconductor makers today.

“It’s highly niche; possibly radical volition ne'er cognize astir it,” said Jha, speaking of the artillery and spot manufacturing marketplace that his institution aims to serve. “But it’s similar down the scenes, this is powering everything… we conscionable don’t spot it successful the regular world.” 

“Every merchandise astir you, similar glasses oregon shoes, slippers, wallets, and headphones — these processes interaction each azygous portion of it,” added Jha.

Niobia’s agents tally connected open-source models that the institution trains further with artillery and spot probe and different data. That grooming is what lets them place pinhole-sized defects. Jha said that Niobia chose open-source models mostly due to the fact that its customers don’t privation their highly delicate institution accusation going into a nationalist AI unreality wherever it mightiness beryllium utilized to bid the model.

“None of our customers privation their proprietary information to spell to the cloud, adjacent if we archer them it’s safe,” said Jha. The startup besides outlines however it approaches cybersecurity, fixed the high-stakes contention among adversarial nations to innovate successful the artillery and spot space. Niobia for lawsuit lets its customers hold their ain encryption keys truthful that the startup cannot decrypt the customers’ data.

Jha said that the company’s precedence for present is to absorption connected 3 specialties, specified arsenic movie coating, electroplating (which is coating a aboveground with metallic utilizing an electrical current), and fluidized furniture absorption (a chemic process wherever coagulated particles are suspended successful a state oregon liquid).

He said that the institution besides wants its exertion to go much autonomous, with the eventual purpose of gathering retired a afloat self-driving mill for its customers.

If you’re funny to larn more, Niobia volition contiguous connected signifier astatine TechCrunch Disrupt this day successful San Francisco. Join us to drawback the institution and the 19 startups against which it’s competing arsenic portion of our storied Startup Battlefield competition, happening today, Tuesday, done Thursday, October 15.