Karttikeya
Mangalam

There is a moment in the UPSC examination hall - India's most punishing competitive test, the one that filters 1.3 million hopefuls down to a few thousand civil servants - when time becomes the real opponent. Candidates get two hours. In June 2024, an AI built by a 28-year-old founder from Muzaffarpur, Bihar completed the entire exam in under seven minutes. Final score: 175 out of 200. The highest the exam has ever seen.

That AI is PadhAI, a product of SigIQ.ai. The founder is Karttikeya Mangalam - PhD in AI from UC Berkeley, alumnus of IIT Kanpur, Meta AI veteran, co-author of 40 peer-reviewed papers, and now the CEO of a startup that wants to drive the cost of elite education down to near zero.

The ambition is not abstract. Mangalam grew up in Muzaffarpur - a Tier-2 city in Bihar where a good tutor is expensive, English-medium schools are a luxury, and the gap between small-town India and the IITs feels oceanic. He crossed that gap through sheer academic force: summa cum laude from IIT Kanpur, a research year at EPFL, an accepted offer at Stanford (later dropped), and finally, a PhD from Berkeley's AI Research Lab (BAIR), advised by computer vision pioneer Prof. Jitendra Malik.

But the credential trail is a distraction. What matters is what Mangalam did with it. In July 2023, months before finishing his PhD, he co-founded SigIQ.ai with Prof. Kurt Keutzer - a distinguished Berkeley professor with 250+ papers and six published books. The thesis was blunt: Benjamin Bloom proved in 1984 that one-on-one tutoring outperforms classroom instruction by two standard deviations. The only problem was cost. Human tutors charge hundreds of dollars per hour. Most families cannot afford that. AI computation, on the other hand, gets cheaper every month.

Mangalam's formula: replace the tutor's labor with AI inference costs. Keep the quality. Drop the price by 99%.

Before SigIQ went public in April 2025, the team spent 18 months in stealth, building and testing in silence. The UPSC result was not a marketing stunt - it was a proof-of-concept. Mangalam's logic: you cannot teach an exam you cannot pass. So they built an AI that passed it better than any human in recorded history. That benchmark is now the floor, not the ceiling.

PadhAI, their UPSC preparation platform, pulled in 200,000 users in six months without a press release. EverTutor.ai, their GRE prep product for the US market, hit 10,000 users in its first three months. The numbers did not come from a viral campaign. They came from students sharing results.

The April 2025 seed round - $9.5 million, co-led by The House Fund and GSV Ventures - came with names that signal something larger than EdTech cheerleading. Duolingo invested. General Catalyst India and Peak XV Partners (formerly Sequoia India) came in. Andy Konwinski, co-founder of Perplexity, wrote a check. Christian Storm, who co-founded Turnitin, did too. This is a group of people who know what it looks like when a software product starts changing behavior at scale.

Mangalam's research career is its own story. Two years at Meta AI, collaborating with Dr. Christoph Feichtenhofer on video understanding and vision transformers. Research stints at MILA (working alongside Yoshua Bengio, one of deep learning's three founding fathers), EPFL, the University of Tokyo, Toyota Research Institute, and the University of Warwick. Two oral presentations at CVPR 2022 - Reversible Vision Transformers and MeMViT - a distinction reserved for fewer than 5% of accepted papers. Work spanning computer vision, NLP, speech, and robotics.

None of that prepared him for building a company. Which is, of course, the point. The research made him certain the technology worked. The company is the part where you have to convince everyone else.

The aspiration inside SigIQ is explicit: make the highest-quality education accessible to anyone, anywhere, at a fraction of the cost. Not as a slogan on a deck slide, but as an operating constraint. Every product decision runs through that filter. If it makes the platform more expensive or less accessible, it does not ship.

Mangalam is 28, building in Berkeley, with offices also in Gurgaon. He likes board games, philosophy of science, and arguments about rationality and AI ethics - the kind of intellectual habits that suggest someone who thinks in systems rather than transactions. Forbes named him to its 30 Under 30 Asia 2025 list in Consumer & Enterprise Technology. KAUST named him an AI Rising Star in 2022. He received the Honda YES Scholar award in 2017, while still an undergraduate at IIT Kanpur.

The trajectory reads as if it were designed backward from a conclusion. But trajectories always look that way in retrospect. What it actually took was growing up in a city where the gap felt impossible, then crossing it, then deciding to close it for everyone who comes after.