A factory that inspects itself
Ask most people what stands between a manufacturing line and a perfect product, and they will picture a person in a high-visibility vest, squinting at a part on a conveyor belt. Kemal Levi has spent his career trying to retire that image. As founder and CEO of Relimetrics, he builds software that lets a factory teach machines to catch their own flaws - no PhD required to operate it, even though one built it.
The company's flagship is ReliVision, an industrial AI platform that Relimetrics describes as inspection-native. The pitch is unusually literal: instead of bolting a camera onto a line and hoping the lighting cooperates, ReliVision lets an organization configure inspection pipelines, train AI models, push them to production, and maintain them - all without writing code. The newest versions lean on generative AI to do something that sounds like science fiction and is closer to plumbing. ReliVision can generate synthetic defects, manufacturing the very flaws it wants a model to recognize, so the system learns to catch problems it has never physically seen.
That idea - that perception itself can be programmed and continuously re-programmed - is the thread that runs through everything Levi has done. He frames the goal as making inspection software-defined, with built-in traceability and auditability, so a vision system keeps evolving as products, environments, and requirements change. In an industry where a single retooled part can break a brittle inspection rig, that adaptability is the whole game.
From thin films to thick steel
Levi did not arrive at the shop floor through robotics or factory automation. He arrived through materials science. He completed his Ph.D. in Materials Science Engineering at Stanford University in 2009, working in the Dauskardt Group on non-destructive testing and imaging applied to polymer thin-film structures - the delicate guts of fuel-cell membranes and solar cells. The work was about reliability at the microscopic scale: how do you measure whether a material will fail without breaking it to find out?
It is a deceptively portable question. Before Relimetrics, Levi took the same toolkit somewhere few materials scientists go: cosmetics and plastic surgery. He pioneered digital image correlation - a technique for tracking how a surface deforms under load - to evaluate treatment efficacy, and his methods for characterizing skin under stress were adopted across the cosmeceutical and medical device sectors. The through-line is hard to miss. Whether the surface is a solar cell, human skin, or an automotive module, Levi keeps asking the same thing: what is this material doing, and can a camera tell me before it is too late?
Program perception itself - so the vision system keeps evolving as the product, the environment, and the requirements change.
Two headquarters, one obsession
Levi founded Relimetrics in 2013, simultaneously in Silicon Valley and Berlin, alongside co-founder Burak Acar, who also holds a PhD. Running a company across the Pacific-to-Europe spread is its own kind of stress test. The split lets Relimetrics serve clients across the Americas, Asia, and Europe, and it has grown into an interdisciplinary team of engineers, designers, and AI specialists - more than two dozen at last count - rather than a single-location shop.
The customer list reads like a tour of heavy industry: automotive module inspection, wind-turbine blade inspection, construction-panel checks, X-ray and ultrasound non-destructive testing, surface inspection with line-scan cameras. Relimetrics positions its software as hardware-agnostic and deployable at the edge or in the cloud, which is the unglamorous detail that actually matters when a manufacturer already owns a building full of expensive cameras it does not want to throw away.
The serial entrepreneur's patience
Founded in 2013, Relimetrics has been chasing the same idea for more than a decade - a long horizon in a field where AI fashions change every eighteen months. Levi is a patent holder and published researcher, and a frequent speaker on non-destructive testing, computer vision, and mechanical reliability at venues like IoT Slam and the Manufacturing IT & OT Summit. The company is a member of the NVIDIA Inception program and has raised Series A funding from a roster that includes Brighter Capital, Merus Capital, Newfund Capital, and Quest Venture Partners.
What makes the arc interesting is not the funding rounds. It is the refusal to pivot. When the generative-AI wave arrived, plenty of computer-vision companies repositioned overnight. Levi's response was to fold GenAI into a thesis he had already been building for years - synthetic defects, adaptive inspection logic, continuously evolving models - rather than chase a new one. The tools changed. The question he is asking did not.
Why it matters
Quality assurance is the part of manufacturing nobody photographs for the brochure, and yet it decides whether a product is safe, whether a recall happens, whether a battery pack or a turbine blade does what it promised. Levi's bet is that this work should be code: versioned, auditable, retrainable on a Tuesday afternoon by someone who is not an AI engineer. If he is right, the person in the high-visibility vest does not disappear. They just stop squinting, and start supervising a machine that learned to see.