Niobia manufacturing AI explains defect causes in minutes

Niobia’s system puts small AI agents right on the production equipment. Each one runs on its own self-contained GPU (a graphics chip that does the heavy math for AI) bolted to the machine, so it behaves a lot like a sensor that watches every part go by.

Those agents look for defects, including flaws as tiny as pinholes and scratches. When they find one, the system doesn’t just raise an alarm. It pulls the data from the line into one place, and engineers can then type questions into a chatbot to work out what caused the problem and keep digging until they reach the real fault.

That second step is the point. Spotting a flaw is only half the job. The slow, costly part has always been figuring out which setting, chemical or machine went wrong, and that detective work is what Niobia is trying to shrink from days to minutes.

New chips and battery designs rarely work perfectly the first time. Problems crop up during research and development, and when they do, engineers can spend days or longer finding a fix before the line restarts. During that time nothing useful comes out of a very expensive factory.

Single mistakes can run into the millions of dollars and push back a whole product schedule. When the same parts are also scarce, a stalled line has knock-on effects for every company waiting on those parts, from phone makers to carmakers.

Niobia was started by Gaurav Jha and Harveen Kaur. Both hold chemistry PhDs and have worked inside chipmakers and battery makers, so they have seen these delays from the factory side.

Jha admits the work is specialized. He described it as “extremely niche” but said it quietly sits behind a huge range of everyday things, from headphones and glasses to wallets and shoes. He also says Niobia serves a market worth billions of dollars that includes some of today’s largest battery and semiconductor makers.

For now the company is concentrating on three manufacturing steps:

• Electroplating: using an electric current to coat a surface with a thin layer of metal.
• Film coating: laying down thin films of material on a part.
• Fluidized bed reaction: a chemical process where solid particles float in a flowing gas or liquid so they react evenly.

Longer term, Jha wants the technology to run more of the line on its own, with the end goal of a fully self-driving factory for customers.

Factory data is some of the most guarded information a manufacturer has. It reveals recipes, yields and the problems a company hasn’t solved yet. Niobia’s customers told it plainly that they don’t want any of that sent to a public cloud, even with promises that it would be safe.

So Niobia builds its agents on open source AI models and trains them further on battery and chip research and other data. That extra training is what teaches them to recognize pinhole-sized flaws. Open models also mean the system doesn’t depend on a big AI provider’s servers, where customer data might be used to train someone else’s model.

Security goes a step further. Customers hold their own encryption keys, which means Niobia itself can’t unlock and read their data. That matters in an industry where countries are racing each other on chip and battery technology and industrial spying is a real worry.

For a manufacturer weighing the tool, that design answers the first question any security team will ask: who else can see our production data? With Niobia, the answer is meant to be nobody outside the company.

You will never see Niobia’s name on a box. But if tools like this work at scale, the effects show up in ordinary ways:

• Fewer delays: lines that restart in hours instead of days can keep new phones, laptops and EVs closer to their launch dates.
• Less waste: catching a flaw early means fewer bad parts and less scrapped material.
• Steadier supply: when chips and batteries are scarce, every extra day of production helps ease shortages.

Keep expectations grounded, though. Niobia is an early-stage startup that is still raising money, and it hasn’t published customer names or hard numbers on how much time it saves. The minutes-not-days claim is its goal, not an independent result.

Niobia pitched on stage in San Francisco in mid-October as one of 20 startups picked for the Startup Battlefield competition at TechCrunch Disrupt 2026. Each team gave a live demo to investors and tech leaders.

It didn’t make the final five. The finalists were Apate, Flox Intelligence, Industrial Next, Intero Biosystems and Satlyt, and the $100,000 cup went to Intero Biosystems, a Denver startup that grows tiny lab versions of human gut tissue for drug testing. Australian scam-fighting startup Apate came second.

Even so, a top-20 slot puts Niobia in front of investors, and the company lists itself as currently raising.