How to Turn Academic Papers into Successful Startups

How to Turn Academic Papers into Successful Startups

What are some ways to approach the deep tech space?

“I think going back to it, it comes down to reading these papers, figuring out what can be productionized, how to write production-level code, but also having some sort of business sense of how to take it to the market. That's just how a lot of these deep tech companies are born. Can you ingest those papers, understand how to read them, and try to build something off it?”

How would you advise a software engineering student in India to self-study AI?

That's a really good question. It's a tricky question. I can talk about my journey, and it's similar to yours. Step one is to find a niche that you're really interested in. For me, it was neuroscience, and I started reading everything about neuroscience, especially neuroscience and AI. I was super into it, so I was reading all the general stuff that was out there because these are easier to understand. YC has a lot of internal things that we're putting out. Google has a lot of cool internal white papers on neuroscience. I started writing some of my own on neuroscience as well.”

How do you manage reading and understanding complex academic papers?

“It takes a long time to read. Even if you read 20 papers, you would still feel like it's never easy. So you want to go section by section. You want to tackle the section, fully understand it, and if you read it a few times, you kind of get it. Now, with OpenAI, I often just copy-paste a paragraph and open it up to give me a summary with ChatGPT and then go into the depths of it. I really leverage ChatGPT to understand what's going on, and then go section by section.”

Did you need to learn foundational biology for your neuroscience work?

“Yes, I did. I spent about three to four months on it. I knew I wanted to work with brain waves. I found brain waves fascinating for a long time. My co-founder had been writing papers on this for 10 years, so he knew everything about it. I knew what my niche was; it wasn't the brain in general, but brain applications using EEG. I learned everything about it that I possibly could. I bought a few books on it, advised by our advisors, some of the top epileptologists. You start with easy 'for dummies' kind of books and then build up from there.”

How did you become interested in using EEG for brain applications?

“At the Google Moonshot Factory for Google X, where they work on weird things, there was a cool project that spun out into a company called Next Sense. They were building earphones where you can listen to music but also track EEG of your brain via the earphones. I was talking to some engineers, and the applications can be crazy to the point where right now neuroscience is one of the few fields that is subjective. For mental health or ADHD, you fill out a form, they track your eye movement, but it's all subjective, on the doctor's discretion.”

How did you come across your co-founder Dimitri?

“I met my co-founder Dimitri through mutual friends. He did his PhD in hospitals, and I was looking for someone specific in neuroscience. He was basically like, 'I know everything about this problem.' He'd been working on EEG for many years.”

What was your process for reading papers and determining their value?

“The way I would always go about it is leveraging ChatGPT. I put a bunch of stuff in there, read the summaries, and decide whether it is worth it or not. ChatGPT helped me with a bunch of things, like understanding the factor rating, how much it has been quoted, and if it is high value. That helped me determine the right papers to read.”

How did you gauge the commercial potential of EEG models?

“It was really about having an understanding of how language foundation models work and thinking, 'Why can't this be applied to this thing?' Papers kept coming out back-to-back, so it was a positive signal. They were all slightly improving on each other, and the missing link was that it was in the lab, so they didn't have enough money to burn on compute.”

What was your approach to researching the latest developments in neuroscience and AI?

“It depends. A lot of times, I became a part of neuroscience groups and AI communities, and I would just read what was circulating in them. There are Slack channels where we discuss new and interesting papers and developments.”

How did you stay updated with the latest papers and trends?

“There’s a group called Braingels, a prominent angel investing group with a lot of neuroscience people who have built and sold companies. The founders of Control Labs and Inflection AI are in it. These guys scout for the next big startup, and they share interesting developments that could be commercialized.”

How did you know which academic papers to focus on?

“The highly cited review articles from prominent neuroscience labs or academic papers are the ones to focus on. The internet algorithms help too. Once it finds out you're in the neuroscience community, you see the latest papers on LinkedIn and other platforms.”

How did you determine the funding needed for EEG model development?

“We looked at the basic outcomes and the open-source data being used, which depended on around $100,000 in compute. We discussed this with my co-founder, who is an AI scientist, confirmed our thoughts, and estimated that if $100,000 got to a certain point, raising $5-6 million could take it much further.

Link to transcript

What are some ways to approach the deep tech space?

“I think going back to it, it comes down to reading these papers, figuring out what can be productionized, how to write production-level code, but also having some sort of business sense of how to take it to the market. That's just how a lot of these deep tech companies are born. Can you ingest those papers, understand how to read them, and try to build something off it?”

How would you advise a software engineering student in India to self-study AI?

That's a really good question. It's a tricky question. I can talk about my journey, and it's similar to yours. Step one is to find a niche that you're really interested in. For me, it was neuroscience, and I started reading everything about neuroscience, especially neuroscience and AI. I was super into it, so I was reading all the general stuff that was out there because these are easier to understand. YC has a lot of internal things that we're putting out. Google has a lot of cool internal white papers on neuroscience. I started writing some of my own on neuroscience as well.”

How do you manage reading and understanding complex academic papers?

“It takes a long time to read. Even if you read 20 papers, you would still feel like it's never easy. So you want to go section by section. You want to tackle the section, fully understand it, and if you read it a few times, you kind of get it. Now, with OpenAI, I often just copy-paste a paragraph and open it up to give me a summary with ChatGPT and then go into the depths of it. I really leverage ChatGPT to understand what's going on, and then go section by section.”

Did you need to learn foundational biology for your neuroscience work?

“Yes, I did. I spent about three to four months on it. I knew I wanted to work with brain waves. I found brain waves fascinating for a long time. My co-founder had been writing papers on this for 10 years, so he knew everything about it. I knew what my niche was; it wasn't the brain in general, but brain applications using EEG. I learned everything about it that I possibly could. I bought a few books on it, advised by our advisors, some of the top epileptologists. You start with easy 'for dummies' kind of books and then build up from there.”

How did you become interested in using EEG for brain applications?

“At the Google Moonshot Factory for Google X, where they work on weird things, there was a cool project that spun out into a company called Next Sense. They were building earphones where you can listen to music but also track EEG of your brain via the earphones. I was talking to some engineers, and the applications can be crazy to the point where right now neuroscience is one of the few fields that is subjective. For mental health or ADHD, you fill out a form, they track your eye movement, but it's all subjective, on the doctor's discretion.”

How did you come across your co-founder Dimitri?

“I met my co-founder Dimitri through mutual friends. He did his PhD in hospitals, and I was looking for someone specific in neuroscience. He was basically like, 'I know everything about this problem.' He'd been working on EEG for many years.”

What was your process for reading papers and determining their value?

“The way I would always go about it is leveraging ChatGPT. I put a bunch of stuff in there, read the summaries, and decide whether it is worth it or not. ChatGPT helped me with a bunch of things, like understanding the factor rating, how much it has been quoted, and if it is high value. That helped me determine the right papers to read.”

How did you gauge the commercial potential of EEG models?

“It was really about having an understanding of how language foundation models work and thinking, 'Why can't this be applied to this thing?' Papers kept coming out back-to-back, so it was a positive signal. They were all slightly improving on each other, and the missing link was that it was in the lab, so they didn't have enough money to burn on compute.”

What was your approach to researching the latest developments in neuroscience and AI?

“It depends. A lot of times, I became a part of neuroscience groups and AI communities, and I would just read what was circulating in them. There are Slack channels where we discuss new and interesting papers and developments.”

How did you stay updated with the latest papers and trends?

“There’s a group called Braingels, a prominent angel investing group with a lot of neuroscience people who have built and sold companies. The founders of Control Labs and Inflection AI are in it. These guys scout for the next big startup, and they share interesting developments that could be commercialized.”

How did you know which academic papers to focus on?

“The highly cited review articles from prominent neuroscience labs or academic papers are the ones to focus on. The internet algorithms help too. Once it finds out you're in the neuroscience community, you see the latest papers on LinkedIn and other platforms.”

How did you determine the funding needed for EEG model development?

“We looked at the basic outcomes and the open-source data being used, which depended on around $100,000 in compute. We discussed this with my co-founder, who is an AI scientist, confirmed our thoughts, and estimated that if $100,000 got to a certain point, raising $5-6 million could take it much further.

Link to transcript