Founder · Professor · Builder
Reza Zadeh
He teaches Stanford PhDs in the morning and ships AI to aerospace factories in the afternoon. The throughline: teaching machines to see.
The professor who'd rather ship
He teaches Stanford PhDs in the morning and ships AI to aerospace factories in the afternoon. The throughline: teaching machines to see.
Walk into a factory running Matroid and a worker can train an AI to spot a missing bolt before lunch - no code, no PhD, no data-science team.
That product is the whole point of Reza Zadeh. He spent years deep inside the machinery of large-scale machine learning - the distributed math, the graph algorithms, the libraries that power thousands of clusters - and then turned around and tried to make all of it disappear behind a button. Matroid, the company he founded in 2016, lets people build and deploy computer-vision detectors without writing a line of code. Point it at a video stream. Tell it what to watch for. It watches.
The ambition behind it is almost absurdly large. Zadeh describes Matroid's mission in two words: detecting everything. He knows that finishing that sentence would require something close to full artificial intelligence, which he openly admits is decades away. He is building toward it anyway.
Twitter's first machine-learning product. The math inside Apache Spark. The founding team at Databricks. He keeps arriving early.
Before Matroid, Zadeh built the algorithm behind Twitter's Who To Follow - the recommendation engine that suggests accounts to new users. It was the first product at the company to use machine learning, and he later released the work to open source. If you ever joined Twitter and immediately got a list of people worth following, you brushed against his code.
Around the same time, he became a co-author of Apache Spark, the cluster-computing framework that reshaped how the industry processes data at scale. He wrote its linear algebra package and co-built MLlib, its machine-learning library. That code now runs quietly inside industrial and academic computing environments around the world - the kind of infrastructure most people never see but constantly depend on. He was also a founding team member at Databricks, the company built to commercialize Spark.
He did all of this while finishing a Stanford PhD in computational mathematics under Gunnar Carlsson, a pioneer of topological data analysis. His thesis, "Large Scale Graph Completion," won the Gene Golub Outstanding Thesis Award. He also picked up a KDD Best Paper Award along the way. Then, instead of choosing between research and building, he refused to choose. He stayed at Stanford as an adjunct professor, teaching two PhD-level courses - Distributed Algorithms and Optimization, and Discrete Mathematics and Algorithms - and founded a company at the same time.
It is a very exciting time to be in computer vision because machine learning has completely overturned it. I've never seen such a big impact of machine learning in any field, and it is building on itself rapidly as we detect more and more things.- Reza Zadeh
That refusal to pick a lane is the most consistent thing about him. The same person grading distributed-systems problem sets signs off on AI deployed in real factories. The researcher who once led work tracking earthquake damage through machine learning - a project that drew wide media attention as an example of real-time social information flowing faster than official channels - is the same operator pitching enterprise customers on defect detection.
Zadeh was born during the Iran-Iraq war, in the under-siege city of Ahvaz. His family left for London, where he grew up until he was 17, then moved again to Toronto. He took his undergraduate degree at the University of Waterloo, a master's at Carnegie Mellon, and his PhD at Stanford - all in computer science and mathematics.
He didn't wait for the credentials to start working. At 18 he was already on the Google Research team. By the time most people finish a doctorate, he'd helped build production machine learning at two of the most influential tech companies of the decade.
Plenty of founders treat a professorship as a line on a bio, a title they keep but no longer earn. Zadeh kept teaching the hard stuff. His two Stanford courses - Distributed Algorithms and Optimization, and Discrete Mathematics and Algorithms - are PhD-level, the kind that sort out who actually wants to do the work. He has been recognized as a best instructor, which is not a thing that happens by coasting.
The reason the two halves of his life feed each other is simple. Computer vision in 2026 is not a settled engineering discipline. It is a moving research frontier, and the gap between a paper and a product can be measured in weeks. A founder who can read the latest distributed-training work and also ship it to a customer has an edge that a pure operator can't buy. Zadeh spent his twenties building exactly that muscle - turning theory into infrastructure that other people quietly run.
It shows up in how Matroid is built. The premise is that the deepest expertise should be invisible to the user. All the distributed math, the model training, the optimization tricks he could lecture on for an hour - the product hides every bit of it. A quality inspector on a manufacturing line never sees a neural network. They see a detector that flags the part that's wrong. That inversion, putting the hardest engineering at the bottom and the simplest experience on top, is the same instinct that made MLlib useful: give people powerful tools without making them earn a degree first.
For all verbs to be detected, eventually we will need full AI - which I don't think will come around for many decades. But probably in my lifetime. Reza Zadeh, on Matroid's mission to detect everything
Most founders pick a problem they can solve. Zadeh picked one he can't - at least not yet. "Detecting everything" is deliberately unfinishable, a horizon he's chosen precisely because it never arrives. Every new thing a camera learns to recognize moves the line forward and reveals how far it still is.
His read on hardware is just as long-range. "Eventually all cameras will have some ability to understand what they're looking at," he has said - a future where vision isn't a feature you add but a default the world ships with. Matroid is his attempt to be early to that, the way he was early to Spark, early to Databricks, early to machine learning at Twitter.
In 2024 the work pulled him into unlikely rooms. He met Pope Francis at the G7 to talk about deploying AI responsibly, and King Charles III at Buckingham Palace to talk about Matroid in aerospace factories and the King's sustainability push for space. Two heads of very different institutions, one year, one subject: what happens when machines start to see.
Eventually all cameras will have some ability to understand what they're looking at.- Reza Zadeh
It was an honor to meet King Charles III at Buckingham Palace. We spoke about Matroid deployments in aerospace factories.- Reza Zadeh, 2024