Why We Invested in Shelfmark

By: Tim Streit

Walk through almost any factory making rolled goods, fabric, labels, industrial film, webbing, and you’ll find the same quality control system: a person standing next to a production line moving hundreds of feet a minute, looking at it. That has been the state of the art in continuous flow manufacturing quality control for decades. Not because manufacturers haven’t tried to do better. Because until recently, nothing better existed.

We’re proud to partner with Pat O’Donnell and Craig Markovitz, co-founders of Shelfmark, who are building the AI inspection platform that continuous-flow manufacturing has been waiting for.

Legacy machine vision players like Keyence, Cognex, and ISRA were built decades ago for discrete, static products. Their systems rely on rules-based algorithms that require controlled lighting, fixed geometry, and tightly scripted defect libraries. This type of technology hasn’t transferred well to continuous flow manufacturing, where lines run at extremely high speeds.

The result: continuous-flow manufacturers, the companies making the fabric in your clothes, the labels on your prescriptions, the material in military parachutes, were left out. We spoke with a 30-year industry veteran who put it plainly: the incumbents won’t bid on a product if they don’t already have the algorithm for it. And they don’t have algorithms for continuous flow.

That’s the gap Shelfmark fills.

Why now? Advances in computer vision and sensor technology have unlocked data capture at a resolution and scale that was previously infeasible. Paired with modern AI, this enables dynamic modeling, pattern recognition, and near-real-time detection and resolution.

Shelfmark’s platform pairs deep-learning computer vision with line-scan camera hardware and IoT sensors to inspect every inch of material at full production speed. No sampling, no spot checks, no hoping the defect happened to be in the section someone glanced at.

What makes this compound over time is what happens after detection. Shelfmark generates a “roll map report card” for every production run, a structured quality record tied to that specific roll of material. That record is beginning to travel up and down supply chains. Customers are pushing Shelfmark on their own suppliers. We’ve seen a lot of go-to-market strategies; ‘your customers do it for you’ is one of the best ones.

One Fortune 500 customer built their own internal ROI model, presented to their CEO to justify rolling Shelfmark out globally, showing $40 million in five-year savings net of cost. Usually the optimistic spreadsheet is the VC’s job. This time the customer did it for us.

Twenty minutes into my first call with Pat, the CEO, it was obvious he hadn’t built a tool and gone looking for a problem. He’d spent months watching the problem happen and built what the people would buy.

Before Pat wrote a single line of code, he spent months on factory floors understanding this market firsthand. He rode alongside workers on early shifts at factories and production floors, absorbing what the job actually looked like before designing software to change it. His background spans industrial engineering, Deloitte consulting, and software product studios. He speaks the language of plant managers and the language of engineers, which matters in a market where the buyer is a floor operations lead, not a CTO.

Co-founder Craig Markovitz previously founded Blue Belt Technologies, a CMU Robotics Institute spinout in surgical robotics, acquired by Smith & Nephew. He’s taught entrepreneurship at CMU for a decade, which means he’s seen more startup pitches than almost anyone alive. Shelfmark is the only one he joined.

Physical hardware on production lines creates switching costs that no software-only competitor can bypass without replacing line equipment. Shelfmark has patents filed covering vertical-specific inspection methods. Each deployment generates labeled defect data that makes models more accurate and shortens time-to-value for the next customer in the same vertical.

And as the roll map report card becomes a shared quality standard, pushed upstream and downstream by customers, the company earns a structural position in industrial infrastructure that doesn’t come from a product roadmap. It compounds from deployment.

Shelfmark is operating in a very green space. The incumbents will eventually notice the market they’ve been ignoring — they usually do, once someone else proves it’s worth the trouble. Shelfmark’s head start is leading in deployments – what compounds and accelerates is the data they will generate and analyze, and the network effects of manufacturers, vendors, and customers.

At Grand Ventures, we look for companies quietly becoming mission-critical in large, overlooked markets. A significant segment of global manufacturing is still doing quality control by eyeball in 2026, not for lack of trying, but for lack of a product that worked. Shelfmark built the one that does.

We’re proud to welcome them to the Grand Ventures portfolio.

Extra icing on the cake…

1. My first summer internship nearly 30 years ago was on the plant floor at Ford Motor Company (I studied ME at U of M) and observed quality control first hand…manually and painstakingly.
2. Pittsburgh has been on my radar for over 10 years, waiting for the right company. The engineering and computer vision expertise at CMU is unparalleled.
3. Kudos to Jacqueline for sourcing this opportunity, and Anthony at Armory Square for sharing!
4. We’ve been spending more time on Physical AI the past 6-12 months. This is the product of the early work.
5. I love it when a plan comes full circle! If you go back to my Ford days/engineering undergrad… it’s been 30 years in the making!