Mukul Dhankhar

At 67% of MLB stadiums, when a fan grabs a hot dog and a beer and holds them over a counter, cameras built by Mukul Dhankhar's company create a real-time 3D model of those items - accurate to the millimeter - and ring them up before the seventh-inning stretch becomes a seventh-inning wait. No scanning. No barcodes. No cashier needed. Transaction complete in seven seconds.

Dhankhar co-founded Mashgin in 2013-2014 with Abhinai Srivastava, bringing a specific obsession: that 3D computer vision - the kind he spent years applying to humanoid robots at Toyota and immersive video conferencing at Bell Labs - was being massively underused in everyday commercial settings. The checkout counter, untouched by serious computer science for decades, was the obvious place to start.

What set Dhankhar apart wasn't the idea of faster checkout. Plenty of companies had tried that. It was the insistence on 3D rather than 2D scanning. Most checkout systems - even modern ones - are fundamentally just fancy barcode readers, dependent on a label being in the right orientation. Mashgin's approach builds a volumetric model of whatever lands on the counter. The system doesn't care which way the bag faces. It sees the object the way a person does - in three dimensions - and identifies it on the spot.

The path to that billion started unconventionally. Dhankhar studied Mathematics and Computer Applications at IIT Delhi - one of India's most competitive institutions - then spent years building machine vision systems that most people never saw. At Toyota's humanoid robotics lab, he was teaching a robot how to perceive its environment: walls, objects, obstacles, depth. The robot needed to understand space the way a person does, not the way a 2D camera does. That required building systems that reconstructed three-dimensional reality from sensor data in real time. It's the same fundamental challenge as identifying a bag of Doritos from any angle on a checkout counter. Dhankhar just applied it to a different problem.

Bell Labs came next - two years building vision algorithms for immersive video conferencing, another domain where getting spatial relationships right was the entire job. By the time he and Srivastava started Mashgin, Dhankhar had spent the better part of a decade solving 3D vision problems in industries that demanded near-perfect accuracy. That background shows in Mashgin's 99.99% item identification rate - not a rounded number, not marketing, a verified operational metric across billions of real-world scans.

Mashgin entered Y Combinator's Winter 2015 batch with a prototype and an unusual claim: their system could identify items in under 10 seconds, with no barcode, with no alignment required, with higher accuracy than a human cashier. The claim held up. Matrix Partners led a $8.2M Series A in November 2017. The technology worked across packaged goods, fresh produce, and even hot food - the hardest category, the one most checkout systems simply refuse to handle.

The growth that followed wasn't explosive - it was methodical. Mashgin deployed in convenience stores, then hospital cafeterias, then sports venues, then airports. Each new environment proved the system's adaptability. A stadium concession stand and a hospital cafeteria are completely different logistically - but the underlying 3D vision problem is the same. Ring up whatever someone puts in front of the camera, correctly, fast.

In May 2022, NEA led a $62.5M Series B at a $1.5 billion valuation - unicorn status, granted to Mashgin on the strength of actual deployment numbers, not projections. Monthly transactions at the time of funding stood at 3 million. Three years later, that figure was 40 million. The company reportedly operates in the black; no Series C has been announced or needed.

The milestone that arrived in April 2025 - 1 billion total transactions processed since founding - isn't just a round number. It translates to roughly 2,000 years of human waiting time that simply didn't happen. That's the metric Dhankhar seems to care about: not how many deals were signed, but how much friction was removed from how many real interactions between real people and the things they want to buy.

Strip away the press-release language and here's what Mashgin's checkout kiosk does: multiple 3D cameras capture the counter from different angles simultaneously. Dhankhar's algorithms fuse those feeds into a real-time volumetric model of whatever objects are present. The system matches that model against a database of known products - not by reading a barcode, but by recognizing shape, texture, and spatial characteristics. Items are identified in milliseconds. The transaction completes in seven to ten seconds total.

The "hot food recognition" capability is the most technically demanding part. A branded candy bar has a consistent visual signature. A stadium hot dog or a hospital cafeteria entree is maddeningly variable - different sizes, different toppings, different containers depending on who prepared it. Mashgin handles this through what they call "custom item learning" - a capability that allows new menu items to be trained into the system quickly, without requiring months of data collection.

The 99.99% accuracy figure matters in ways that aren't immediately obvious. One missed scan in 10,000 transactions sounds negligible. At 40 million monthly transactions, that's still 4,000 errors per month - errors that translate to lost revenue, inventory discrepancies, or failed audits. Getting to four nines required solving 3D vision problems that 2D barcode systems never had to confront.

Mashgin's footprint as of 2025 spans industries that share one trait: high transaction volume in short windows. A convenience store at rush hour. A stadium concession stand during the seventh inning. An airport terminal between departures. A hospital cafeteria between visiting hours. In each case, the value proposition is identical - move more people through faster, with fewer staff, at higher accuracy.

The Denver Broncos deployment produced a telling data point: a 34% increase in concession sales per game after Mashgin went live. Faster checkout doesn't just improve customer experience - it directly unlocks revenue that was previously lost to long lines. The math is simple: if a fan gives up waiting and goes back to their seat, that transaction never happened. If checkout takes seven seconds instead of two minutes, it happens.

The 67% MLB stadium figure is the one that lands hardest. Baseball stadiums are enormous, complicated venues with dozens of concession points, inconsistent product mixes, and thousands of transactions compressed into a few hours. Getting to two-thirds of the league is a procurement and deployment achievement as much as a technical one.

Dhankhar's trajectory before Mashgin reads like someone who kept finding the same technical problem in different guises. At Toyota's humanoid robotics lab around 2012, the question was: how do you give a robot an accurate model of the physical space around it? Not a flat image, but a real three-dimensional understanding - where surfaces are, how far away objects sit, what can be reached and what can't. The humanoid robot needed this to move safely and interact with its environment. Dhankhar built it.

Bell Labs was the same problem, different context. Immersive video conferencing requires capturing a room - the geometry of it, the people in it, the objects on the table - and reconstructing it in a way that feels spatially coherent to someone watching remotely. Two years writing vision algorithms for that system deepened the same set of skills: multi-camera fusion, real-time 3D reconstruction, spatial accuracy under noisy conditions.

The IIT Delhi foundation wasn't incidental. The integrated M.Tech in Mathematics and Computer Applications is a demanding program that trains students to build mathematical models of complex systems from first principles - not to apply existing frameworks, but to derive them. That training shows in how Mashgin approached the checkout problem: not by adapting existing retail technology, but by building a fundamentally different system from the image sensor up.

The co-founder pairing with Abhinai Srivastava - who came from Yahoo, Microsoft's Bing Knowledge Graph, and Facebook's data science team - split the company's technical leadership cleanly: Dhankhar owns the vision and hardware stack; Srivastava owns the data and machine learning layers. Ten years in, that division appears to have held.