Does It Actually Work? Rahul Tarak's Decade At The Judging Table

Every hackathon ends the same way. After a night or two of building, a team turns a laptop toward a stranger with a scoring sheet and gets a few minutes to prove that the thing on the screen is real. The slides stop mattering. Either the software runs or it doesn't.

Rahul Tarak has spent close to a decade on both sides of that moment, as the student holding the laptop and as the stranger with the sheet. He has built software used by thousands of people, been picked by some of the most selective programs in tech on the strength of his code, and judged student builders in two countries. The question he brings to the table has barely changed. What has changed is who has to answer it. More and more often, it is an AI agent.

Building The Table

As a student at Oakridge International School in Bangalore, Tarak was already a regular on the city's student hackathon circuit, with first-place finishes at events including Vivum, RHOK and Synt Hacks. Then he decided his own school should have one. He started Codefest, an overnight hackathon run by Oakridge's students, and ran and judged its early editions.

The rules are still simple: teams of two to six, 24 hours, three meals and a snack, then the finalists pitch to a panel of judges. "You can't hide behind a slide deck," Tarak said. "Either it runs in front of the judges or it doesn't."

Codefest outgrew him, which is the point of building one. Its sixth edition, in January 2023, drew 127 students in 38 teams from 13 schools across India. The school now calls it India's largest in-person high-school hackathon, and it has invited him back as a guest and a judge for the tenth edition, in February 2027.

Being Judged

At the University of Toronto, where he studied mathematics and computer science, Tarak went back to holding the laptop. At campus hackathons he became friends with Zack Radisic, who would later build a company with him.

"Rahul builds the thing first and argues about it later," Radisic said. "Most people at a hackathon are trying to impress the judges. He was always trying to impress the users."

In 2020, with classes online, Tarak built Modfy, which ran FFmpeg, the open source engine underneath much of the world's video software, inside a browser tab. People could convert and compress video without uploading it anywhere. That September it reached the front page of Hacker News.

The judges that year were strangers, and what they judged was his code. Pioneer, the accelerator founded by investor Daniel Gross, accepted Tarak in October 2020 largely on the strength of his open source work, which developers had starred more than 1,500 times on GitHub. Pioneer says it accepts fewer than 1 percent of applicants. Major League Hacking picked him for its fellowship that fall, and On Deck's ODX program and the Interact fellowship followed in 2022.

Modfy grew into Scalar, a video editor several people could cut at once, and Radisic became its chief technology officer. He later joined Bun, the JavaScript runtime company since acquired by Anthropic. On Scalar, every trim and cut from every editor had to settle into one consistent project without anyone waiting on anyone else. The team made that work with conflict-free replicated data types, or CRDTs.

Scalar is also where Tarak found the limit of the hackathon. Weekend logic says ship the smallest thing that works and fix it later. Some foundations don't allow that. "There is no world in which we should've MVP'd our multiplayer engine," he wrote in a March 2024 retrospective. A weekend can tell you whether an idea is real. It can't tell you whether it will hold. He judges with both in mind.

Back At The Table

In February 2023 he returned to Bangalore to judge TISB Hacks 3.0, a national hackathon for school students at The International School Bangalore. He sat on the panel with Vivek Sunder, the former chief operating officer of Swiggy, the food delivery company that went public in November 2024 at a valuation of about $11 billion. They scored projects on impact potential, code complexity and innovation, criteria that pull against each other. A hard project can solve a problem nobody has. A good idea can hide code that doesn't run yet.

"Some of those kids were better than I was at that age," Tarak said. "They also had better tools."

Since moving to San Francisco he has kept at it, mentoring student teams and judging hackathons in the US. In October 2025 that brought him to the Palace of Fine Arts, a rotunda built for a 1915 world's fair, for Cal Hacks 12.0. The student-run hackathon out of the University of California, Berkeley, bills itself as the world's largest collegiate hackathon. More than 2,000 students competed for $58,100 in prizes.

The Agents Arrive

This time, agents were everywhere. Many of the sponsor challenges asked teams to build AI systems that take actions on their own instead of only answering questions. That makes the judge's oldest question harder. A fluent reply in a chat window is not proof that anything happened.

"A lot of the agent demos were a chat window with a nice coat of paint," Tarak said. "I kept asking the same thing. Did it actually do something, or did it just talk about doing something?"

The teams that stood out showed their work: what the agent touched, what changed, and how a person could check it or undo it. The question he asks at the judging table is about to become his job. Tarak is joining Composio to lead product. The San Francisco company, backed by Lightspeed Venture Partners, lets AI agents act inside the software where work already lives, from Gmail to Salesforce. More than 100,000 developers build on it, and it handles more than 1 billion agent actions a month.

His advice to the students he judges has not changed since his own first hackathon. "Ship it Friday night, show someone on Saturday, fix it on Sunday," he said. "Then put the code where people can see it. That's how I got found."