I live in Mumbai San Francisco now, made the move in April. Currently working on Clio AI - long horizon RL agents that can finish tasks with high accuracy.

A few things about me:

• Grew up in Udaipur, India. Love the city.
• Undergrad from IIT Kanpur.
• Started working at O9 Solutions in supply chain consulting.
• Worked at Urban Company, Impact Guru, Upwards Fintech.
• Cofounded a couple of startups
  • Blubyn: An AI travel chatbot in 2017.
  • Noomi: A community startup in 2021.
• Now, I research and work on RL.
• Currently working on self verifying agents.

Opinions - both weak and strong

• Curiosity wins out in the long run.
• Go deep on things because you like it.
  • This is personal given I can only cover the width if I understand things deeply enough.
  • I enjoy going deep on multiple things. In particular - programming, writing, physics, marketing, chronic conditions. Some more frequent than others.
  • It’s okay if the interest does not result in anything professionally or personally.
• AI is going to give us more agency than we ever had.
  • Many people are not comfortable with agency and do not know what to do with it.
  • Working with agency requires a level of conviction people will be slow to come to terms with. This presents both opportunities and challenges.
  • This fundamentally changes the way most things happen in this world.
• Anti Inductive things have huge alpha.
  • With many things out there, the value diminishes as more people start doing it. Truly transformative changes require original thinking.
  • There is a unique learning opportunity when you apply first principles thinking to various problems out there.
• We should aim to solve harder problems no matter the domain.
  • This is a personal goal. I feel energized solving a challenging problem.
  • It’s okay if the solution is not perfect.
  • It’s okay if you don’t share the same goal.
• I get my serotonin from:
  • Puzzles
  • making things work. And working out how stuff is configured.
  • I play video games to understand how it would have been coded - how do I beat the system? Gives me more fun than the game itself.
  • It’s not just about understanding, it’s the first step to making things work.

Things I built

• Blubyn
• Noomi
• Clio AI

Research

• Cross Domain Content Generation: ArXiv

Things I like

• Reading. Check out my bookshelf.
• Football. Real Madrid.
• Astrophysics.
  • I always liked it. It’s a mix of math, physics and philosophy.
• History.
  • Not the one taught in school, but specific events and multiple explanations of the same.
  • I like to think about why things happened the way they did. What were the trigger points, and figuring out the truth when you have different sources and perspectives describing the same event in different ways. It’s like a puzzle.

Questions

Some questions I have thought about and would love perspectives about

• Why do people have different individual personalities that go away as a collective?
• How do people make decisions about life changes? Things are constant for a while and suddenly they change.
• What would be a good successor to paper? That is: which technology can preserve our learnings for the next 1000 years? No, digital memory that requires a software version to be readable is not the answer.

Other

I maintain a list of bets I am making in public domain. You can check the git history for log of changes.