The Berkeley AI tutor that walked into India's hardest exam and walked out with the highest score in its history. In seven minutes.
1600 Shattuck Ave · Berkeley, California · A Tuesday
It is late afternoon in the upstairs office above a Berkeley sandwich shop, and a graduate student in Lucknow is asking SigIQ's tutor about the Mauryan Empire for the third time today. The tutor answers, then quietly notices that the student has been confusing two centuries since Monday. It rewrites the next question. It does not sigh. It does not glance at the clock. There are thirty-two more people just like her open in tabs around the world right now, and the model running underneath does not particularly care which one it is talking to.
This is what SigIQ.ai built. Not a chatbot that knows things. A tutor that notices.
The equation goes like this. A genuinely good private tutor in the United States costs around forty thousand dollars a year. The number of students on earth who could use one runs into the hundreds of millions. The math has never closed. It has never come close to closing.
Karttikeya Mangalam, a UC Berkeley AI PhD trained by Jitendra Malik and seasoned at Meta's FAIR and Google DeepMind, kept staring at that equation. So did Kurt Keutzer, the Berkeley professor whose previous startup, DeepScale, was bought by Tesla. In July 2023 they did the obvious thing for two people with their CVs - they started a company.
SigIQ.ai was the result. Two offices. Berkeley and Gurgaon. One thesis: if you could make a tutor that was reliable, patient, and as cheap as the GPU bill, the equation would finally close.
The number SigIQ is trying to demolish. Most students never see a tutor like this. SigIQ thinks they should.
Co-led by The House Fund and GSV Ventures, with Duolingo, Peak XV and Calibrate writing checks alongside.
On June 16, 2024, in a ballroom at The Leela in Delhi, SigIQ's tutor sat down to take the UPSC Prelims. The exam is the front door to India's civil service. About 1.3 million people sit for it every year. Most fail. The smart ones train for years. The actual two-hour test had just ended that morning.
PadhAI took it in under seven minutes. Score: 175 out of 200. The highest in the exam's recorded history.
The room reacted the way rooms react when an old certainty cracks. Carefully.
SigIQ runs one model stack and points it at two of the world's most consequential exams. The bet is that if you can hold a learner's attention through UPSC and GRE, you can hold it through anything.
An AI tutor for India's civil service exam. Adaptive testing, gamified study, 200,000+ learners and a 4.9-star rating in its first six months.
Real-time tutor with voice and screen-share. Launched April 7, 2025. Hit Product Hunt's Product of the Day at launch.
Adaptive practice sets, mock tests and personalized feedback for US graduate-school applicants.
Berkeley AI PhD, advised by computer-vision pioneer Jitendra Malik. Previously a research scientist at FAIR and Google DeepMind. 40+ papers across vision, NLP, speech and robotics. 6,000+ citations. Builds the tutor he wishes he'd had.
Professor in the Berkeley AI Research Lab. Six books, 250+ papers, twelve startups launched. His last company, DeepScale, was acquired by Tesla. His current obsession is making language models small, cheap and trustworthy enough to teach.
The seed round closed April 7, 2025. It was co-led by The House Fund and GSV Ventures. The cap table has the obvious EdTech names. It also has Duolingo. Yes, that Duolingo.
Karttikeya Mangalam and Kurt Keutzer found SigIQ in Berkeley.
PadhAI scores 175/200 on UPSC Prelims live at The Leela, Delhi.
PadhAI crosses 200,000 learners with a 4.9-star rating.
Company exits stealth with $9.5M seed. EverTutor Live launches the same day.
33-person team across Berkeley and Gurgaon, expanding into new exam categories.
It is later now, in the office above the sandwich shop. The graduate student in Lucknow has finished her Mauryan question, gotten three follow-ups right in a row, and closed her laptop with the small private satisfaction of having understood something she did not understand this morning. The tutor has moved on. Another thirty-two tabs. None of them know she is on the other end. None of them care.
The economics of education have always been about scarcity. Good teachers cost what good teachers cost. Time is finite. Patience is finite. SigIQ.ai is the bet that one of those constraints - the one about patience and the one about cost - was never an actual constraint. It was just a missing engine.
That engine now lives at 1600 Shattuck Avenue, and on a quiet server farm somewhere, and in the pocket of a student in Lucknow who is studying for a test she might actually pass.
— 30 —