Every time I think I’ve figured out Kartin Wong, he surprises me. He dodges founder stereotypes, refusing to fit the neat tropes so often projected onto technical founders once a surface level of familiarity is gained.
Let’s take some examples. Kartin is an Informatics Olympiad coder but his outfit of choice is a white tank top with a thick cuban link gold chain. Kartin is the co-creator of opML (a revolutionary approach to on-chain artificial intelligence) but his favorite musical artist is Tyga. He’s an enigma, but it’s this unique synthesization of identities that makes him such a captivating founder, one that is uniquely equipped to build ORA, the world’s first verifiable oracle protocol bringing AI and complex compute onchain.
By all accounts, Kartin was an exceptional child, but it’s a label he refuses to claim. Growing up in Canton, China, he won a National Music Competition in 2012 having been voted the best musical act. The prize? Playing guitar for Taylor Swift at a private concert in Hong Kong.
When I asked him about this, Kartin rushed to clarify that he feels like he gamed the system to win. “I was just a popular kid in my hometown so of course everyone was going to vote for me!”
Outside of his music, Kartin was also a competitive programmer, which ultimately led to his first interactions with crypto. He was part of his hometown’s Informatics Olympiad team, where he competed in competitive programming competitions. Here again, Kartin downplays his programming skills. “To be honest, I really wasn’t a great programmer. There are so many genius programmers in China… I’m not one of them.” Funnily enough, it was his early shortcomings as a programmer that led to him first getting involved in crypto.
He explains that his Olympiad coach needed to find something to occupy Kartin during team practices so as to keep him separate from the more advanced students. His coach had just heard about the famous Bitcoin pizza transaction, and thought maybe he could distract Kartin by getting him to play around with Bitcoin during practice. His teacher tasked Kartin with writing a script that could facilitate Bitcoin mining for the team. Recalling the Python script he produced, Kartin explains, “It was so inefficient. It really didn’t make any sense to use Python for that but it was the only language I knew at the time.”
Efficiencies aside, the Python script worked, and after mining his first Bitcoin in 2013 (~$100 at the time), he realized he may be able to turn this into a side business. Ironically, he didn’t actually start a mining business, which would have been the most lucrative path, but instead stood up a website and sold the script to mining operations globally. All together, he thinks he made about $700k from this operation before he reached college. For him, it was an unfathomable amount of money. “My parents weren’t well off so I had never seen this kind of money before. I thought I was super rich. I thought I was a billionaire. I literally thought I’d never have to work again in my life.”
With the money from his script business, he was able to pay his way to the United States where he went to the University of Arizona to study astronomy. Not pressed for a high-paying career, he thought astronomy would be a fun way to spend the rest of his working years. After all, in his mind he was already a crypto billionaire.
You may ask yourself, what does a 19 year old with $700k in his bank account do when he first experiences independent living? He buys a Ferrari, of course. Reflecting on that financial decision, he admits, “I had no idea what I was doing.”
Reality hit fast, as he soon realized that $700k combined with a 19 year-old’s financial judgment was not in fact going to take him to retirement. An early college romance provided an even starker wakeup call. “I had been hanging out with this girl, we went on a couple of dates and then one day, out of the blue, she told me she was pregnant.” An unplanned pregnancy would be a major challenge for any college student, not to mention one that had only stepped in the US for the first time earlier that year.
Kartin and his partner soon got married, bringing a new focus to the rest of his time at the University of Arizona. He made his way through undergrad as a top student, ending up with a sizable offer to work at Google by his senior year. When asked about how he got an engineering role at a top tech firm coming out of a university not known for a competitive CS program, I finally got him to admit, “Ok ok, I’m actually pretty good at programming.”
At Google, Kartin was assigned to the Trusted Hardware team, which gave him his first taste of cryptography. It was not the most prestigious group to join, “Nobody wants to join the Trusted Hardware group. It’s the most useless group at Google.”
The bad news bears of Google worked on hardware like Google Glass and Android devices to ensure that connections were formally verified using cryptography. “It wasn’t interesting to work on. The servers are so strong that no one cares if the cryptography is elegant. As long as it worked it got shipped.” In a callback to his first Python scripts, here too, efficiency was not a focus.
After one year developing his cryptography skills at Google, Kartin next took a role at TikTok working on cloud privacy. He eventually got to the point where he was managing a team of 80 engineers and wielding a multimillion dollar budget, but after a new direction in TikTok’s strategy, his division became less of a focus and he ultimately decided to leave in 2022. It was time to take some time off, survey the tech landscape and decide what his next move would be. He wondered if there had been any interesting stuff happening in crypto since his Bitcoin mining days in 2013…
“I couldn’t believe how much had changed,” Kartin reflects when asked about his initial reaction to getting back into crypto after a ten year hiatus. During his time away, DeFi had been through a full cycle, every celebrity had an NFT profile picture, and memecoins were pumping. “I looked around and there was all of this new stuff, but it was all so boring. Every alt-L1 was an Ethereum knockoff and all of the L2s were carbon copy AWS-chains.”
There was one trend that had caught his attention however. “I noticed that zero knowledge (ZK) proofs and cryptography more generally were way more sophisticated in crypto than in web2. In web2, no one cares about efficiency. They just want to have the proofs but they don’t care about proof size or how fast it can be generated. In web3, efficiency is extremely important and I realized that there was all of this cryptographic innovation happening that I hadn’t been exposed to at Google or TikTok.”
He started digging into this space, trying to identify what application could most benefit from ZK. “I saw the new L2s and L1s and knew they’d need access to an oracle. Up until this point however, ZK proofs weren’t efficient enough to integrate into an oracle, but with Halo 2 and the first Plonky library, the tech was finally there.”
As he dove into this space, he was introduced to his co-founder Xiaohang Yu, a PhD student at Imperial College London who was working on applied cryptography. They were able to hack together an MVP of their ZK-enabled oracle, naming it ORA.
What Kartin and Xiohang built was a new type of decentralized oracle network utilizing ZK proofs. In crypto, oracles allow on-chain and off-chain data to talk to each other. For example, you could have the outcome of a political race (off-chain data) trigger the closure of a bet on a blockchain-based prediction market (on-chain). What was unique about ORA’s design was that by leveraging ZK technology, their oracle was significantly more secure, decentralized, and performant than existing solutions. Instead of relying on staking and slashing mechanisms for security, their oracle used more advanced cryptography to ensure the integrity of the system, bringing significant advantages over existing competitors like Chainlink and The Graph (read more about the architecture in their whitepaper).
While the technical innovation was there, they were initially met with great skepticism. No investor had seen ZK technology applied to oracles before and as such, numerous VCs they pitched thought they were scam artists. Still, Kartin and Xiohang eventually scraped together a small round and went out to hire more engineers to build their team. It was their next hire that would bring a new level of innovation to their platform.
“Whenever I had down time building ORA, I would browse the Ethereum Foundation forum,” Kartin tells me. “I was always looking for new ZK use cases and new implementations for circuits. One day, I saw a post with a super long, confusing title. It was too long long, stuff that wouldn’t make sense to normal people. No one had commented but I recognized that it was RDOC which has to do with super old optimistic fault proofs. I was like, I have to reach out to this person.”
Kartin sent a cold DM and waited for a response. A few days later, he got a response that shocked him, “The post author told me they’d figured out how to enable a smart contract to run a neural network. It was crazy. If it was true I knew it was something I needed to pursue.”
Why was Kartin so excited by this fringe idea on the forum? Seeing the explosion of AI-generated content but a glaring lack of model transparency, he knew on-chain machine learning facilitated through their zkOracle would be the perfect way to make AI less centralized and more open. The problem was that, to-date, the only solution proposed was zero knowledge machine learning (zkML), a prohibitively expensive approach that was not nearly performant enough for real world applications. However, the approach outlined in the forum post, which leveraged optimistic verification methods instead of ZK, would allow ORA to bring AI on-chain in a highly optimized, performant way.
Kartin discovered that the user on the other end of the forum was an engineer named Conway. Conway was a Computer Science PhD Candidate at Tsinghua University who was deep into on-chain AI research. Kartin wanted to collaborate but was hesitant to work too closely with him. “When I got to know Conway I realized he might not be ready for the pace of a venture-backed company. He told me he just wanted to work on his PhD research, not blitzscale a startup. So, before I got him to join ORA I actually referred him to write some research articles for Sequoia China.” It wasn’t a great match.
“Conway went off and spent weeks working on a detailed article on decentralized AI for Sequoia. He turned it in and the people at Sequoia were like, ‘We have no idea what the fuck this guy is talking about.’ They had no interest in working with him. He came back to me after that experience with his tail tucked and I realized we needed to work together.”
Unlike the Sequoia experience, Conway and Kartin were a match made in heaven. “If you look at our publications on AI crypto research (found at ai.ORA.io), Conway did 90% of that work. He’s brilliant. As the founder, I realized I just needed to serve as an adapter, a translator for him. My role was to get out of his way and help hone his innovation and convey his work to the outside world.”
Together, Kartin and Conway were able to bring this new framework to life, dubbing it optimistic machine learning (opML). Once implemented, the opML approach to on-chain machine learning immediately showed a significant improvement over zkML. They couldn’t believe the contrast in hardware requirements for their model versus zkML – opML could run a large language model (e.g. 7B-LLaMA) on a common PC without a GPU. In contrast, zkML would take nearly 80 minutes to generate a proof for a much smaller, 1M-nanoGPT model on the same hardware.
Given this computational efficiency, opML has been able to unlock novel use cases for on chain machine learning like content authenticity, on chain AI games, sensitive data processing (e.g. healthcare), and future use cases they’ve only just scratched the surface of.
Reflecting on my conversation with Kartin, I can’t help but wonder how he was able to so adeptly work with Conway to coax this groundbreaking innovation out of him. How was he able to translate this vision when the world’s leading venture fund couldn’t even make it past one article? I think it has something to do with how Kartin’s myriad identities allow him to connect with people from all walks of life. His technical background allowed him to understand the potential in an overlooked forum post while his soft skills meant that he was also able to take that diamond of an idea and polish it into opML. When I ask him about where he thinks these abilities come from, he takes me back to his childhood.
“Growing up I thought I was going to be a musician. I actually had a contract from a recording label around the time of the Taylor Swift performance. So, I think it’s just in me to perform. I have to perform. And when I perform, I have to figure out my audience and how to connect with them.”
Reflecting on what he’s most excited about with ORA’s future, Kartin tells me there is much more to come with opML. “It’s kind of new. It’s like a newborn baby," he explains. “It just came out this past April so we need to put many more resources into building out the opML library. There are bottlenecks that we just resolved but it’s going to be a bit before we can announce them.”
.jpeg)
By the end of my time with Kartin, I can’t help but share his excitement. Having already built and refined their zkOracle platform, his team has rapidly pushed forward into uncharted territory with opML. This product has the potential to radically transform the visibility and verifiability of artificial intelligence. It’s a big problem, one that necessitates a strong founder who can set a vision in a dynamically shifting landscape. I couldn’t think of anyone better equipped to tackle that challenge than Kartin. ✦
Every time I think I’ve figured out Kartin Wong, he surprises me. He dodges founder stereotypes, refusing to fit the neat tropes so often projected onto technical founders once a surface level of familiarity is gained.
Let’s take some examples. Kartin is an Informatics Olympiad coder but his outfit of choice is a white tank top with a thick cuban link gold chain. Kartin is the co-creator of opML (a revolutionary approach to on-chain artificial intelligence) but his favorite musical artist is Tyga. He’s an enigma, but it’s this unique synthesization of identities that makes him such a captivating founder, one that is uniquely equipped to build ORA, the world’s first verifiable oracle protocol bringing AI and complex compute onchain.
By all accounts, Kartin was an exceptional child, but it’s a label he refuses to claim. Growing up in Canton, China, he won a National Music Competition in 2012 having been voted the best musical act. The prize? Playing guitar for Taylor Swift at a private concert in Hong Kong.
When I asked him about this, Kartin rushed to clarify that he feels like he gamed the system to win. “I was just a popular kid in my hometown so of course everyone was going to vote for me!”
Outside of his music, Kartin was also a competitive programmer, which ultimately led to his first interactions with crypto. He was part of his hometown’s Informatics Olympiad team, where he competed in competitive programming competitions. Here again, Kartin downplays his programming skills. “To be honest, I really wasn’t a great programmer. There are so many genius programmers in China… I’m not one of them.” Funnily enough, it was his early shortcomings as a programmer that led to him first getting involved in crypto.
He explains that his Olympiad coach needed to find something to occupy Kartin during team practices so as to keep him separate from the more advanced students. His coach had just heard about the famous Bitcoin pizza transaction, and thought maybe he could distract Kartin by getting him to play around with Bitcoin during practice. His teacher tasked Kartin with writing a script that could facilitate Bitcoin mining for the team. Recalling the Python script he produced, Kartin explains, “It was so inefficient. It really didn’t make any sense to use Python for that but it was the only language I knew at the time.”
Efficiencies aside, the Python script worked, and after mining his first Bitcoin in 2013 (~$100 at the time), he realized he may be able to turn this into a side business. Ironically, he didn’t actually start a mining business, which would have been the most lucrative path, but instead stood up a website and sold the script to mining operations globally. All together, he thinks he made about $700k from this operation before he reached college. For him, it was an unfathomable amount of money. “My parents weren’t well off so I had never seen this kind of money before. I thought I was super rich. I thought I was a billionaire. I literally thought I’d never have to work again in my life.”
With the money from his script business, he was able to pay his way to the United States where he went to the University of Arizona to study astronomy. Not pressed for a high-paying career, he thought astronomy would be a fun way to spend the rest of his working years. After all, in his mind he was already a crypto billionaire.
You may ask yourself, what does a 19 year old with $700k in his bank account do when he first experiences independent living? He buys a Ferrari, of course. Reflecting on that financial decision, he admits, “I had no idea what I was doing.”
Reality hit fast, as he soon realized that $700k combined with a 19 year-old’s financial judgment was not in fact going to take him to retirement. An early college romance provided an even starker wakeup call. “I had been hanging out with this girl, we went on a couple of dates and then one day, out of the blue, she told me she was pregnant.” An unplanned pregnancy would be a major challenge for any college student, not to mention one that had only stepped in the US for the first time earlier that year.
Kartin and his partner soon got married, bringing a new focus to the rest of his time at the University of Arizona. He made his way through undergrad as a top student, ending up with a sizable offer to work at Google by his senior year. When asked about how he got an engineering role at a top tech firm coming out of a university not known for a competitive CS program, I finally got him to admit, “Ok ok, I’m actually pretty good at programming.”
At Google, Kartin was assigned to the Trusted Hardware team, which gave him his first taste of cryptography. It was not the most prestigious group to join, “Nobody wants to join the Trusted Hardware group. It’s the most useless group at Google.”
The bad news bears of Google worked on hardware like Google Glass and Android devices to ensure that connections were formally verified using cryptography. “It wasn’t interesting to work on. The servers are so strong that no one cares if the cryptography is elegant. As long as it worked it got shipped.” In a callback to his first Python scripts, here too, efficiency was not a focus.
After one year developing his cryptography skills at Google, Kartin next took a role at TikTok working on cloud privacy. He eventually got to the point where he was managing a team of 80 engineers and wielding a multimillion dollar budget, but after a new direction in TikTok’s strategy, his division became less of a focus and he ultimately decided to leave in 2022. It was time to take some time off, survey the tech landscape and decide what his next move would be. He wondered if there had been any interesting stuff happening in crypto since his Bitcoin mining days in 2013…
“I couldn’t believe how much had changed,” Kartin reflects when asked about his initial reaction to getting back into crypto after a ten year hiatus. During his time away, DeFi had been through a full cycle, every celebrity had an NFT profile picture, and memecoins were pumping. “I looked around and there was all of this new stuff, but it was all so boring. Every alt-L1 was an Ethereum knockoff and all of the L2s were carbon copy AWS-chains.”
There was one trend that had caught his attention however. “I noticed that zero knowledge (ZK) proofs and cryptography more generally were way more sophisticated in crypto than in web2. In web2, no one cares about efficiency. They just want to have the proofs but they don’t care about proof size or how fast it can be generated. In web3, efficiency is extremely important and I realized that there was all of this cryptographic innovation happening that I hadn’t been exposed to at Google or TikTok.”
He started digging into this space, trying to identify what application could most benefit from ZK. “I saw the new L2s and L1s and knew they’d need access to an oracle. Up until this point however, ZK proofs weren’t efficient enough to integrate into an oracle, but with Halo 2 and the first Plonky library, the tech was finally there.”
As he dove into this space, he was introduced to his co-founder Xiaohang Yu, a PhD student at Imperial College London who was working on applied cryptography. They were able to hack together an MVP of their ZK-enabled oracle, naming it ORA.
What Kartin and Xiohang built was a new type of decentralized oracle network utilizing ZK proofs. In crypto, oracles allow on-chain and off-chain data to talk to each other. For example, you could have the outcome of a political race (off-chain data) trigger the closure of a bet on a blockchain-based prediction market (on-chain). What was unique about ORA’s design was that by leveraging ZK technology, their oracle was significantly more secure, decentralized, and performant than existing solutions. Instead of relying on staking and slashing mechanisms for security, their oracle used more advanced cryptography to ensure the integrity of the system, bringing significant advantages over existing competitors like Chainlink and The Graph (read more about the architecture in their whitepaper).
While the technical innovation was there, they were initially met with great skepticism. No investor had seen ZK technology applied to oracles before and as such, numerous VCs they pitched thought they were scam artists. Still, Kartin and Xiohang eventually scraped together a small round and went out to hire more engineers to build their team. It was their next hire that would bring a new level of innovation to their platform.
“Whenever I had down time building ORA, I would browse the Ethereum Foundation forum,” Kartin tells me. “I was always looking for new ZK use cases and new implementations for circuits. One day, I saw a post with a super long, confusing title. It was too long long, stuff that wouldn’t make sense to normal people. No one had commented but I recognized that it was RDOC which has to do with super old optimistic fault proofs. I was like, I have to reach out to this person.”
Kartin sent a cold DM and waited for a response. A few days later, he got a response that shocked him, “The post author told me they’d figured out how to enable a smart contract to run a neural network. It was crazy. If it was true I knew it was something I needed to pursue.”
Why was Kartin so excited by this fringe idea on the forum? Seeing the explosion of AI-generated content but a glaring lack of model transparency, he knew on-chain machine learning facilitated through their zkOracle would be the perfect way to make AI less centralized and more open. The problem was that, to-date, the only solution proposed was zero knowledge machine learning (zkML), a prohibitively expensive approach that was not nearly performant enough for real world applications. However, the approach outlined in the forum post, which leveraged optimistic verification methods instead of ZK, would allow ORA to bring AI on-chain in a highly optimized, performant way.
Kartin discovered that the user on the other end of the forum was an engineer named Conway. Conway was a Computer Science PhD Candidate at Tsinghua University who was deep into on-chain AI research. Kartin wanted to collaborate but was hesitant to work too closely with him. “When I got to know Conway I realized he might not be ready for the pace of a venture-backed company. He told me he just wanted to work on his PhD research, not blitzscale a startup. So, before I got him to join ORA I actually referred him to write some research articles for Sequoia China.” It wasn’t a great match.
“Conway went off and spent weeks working on a detailed article on decentralized AI for Sequoia. He turned it in and the people at Sequoia were like, ‘We have no idea what the fuck this guy is talking about.’ They had no interest in working with him. He came back to me after that experience with his tail tucked and I realized we needed to work together.”
Unlike the Sequoia experience, Conway and Kartin were a match made in heaven. “If you look at our publications on AI crypto research (found at ai.ORA.io), Conway did 90% of that work. He’s brilliant. As the founder, I realized I just needed to serve as an adapter, a translator for him. My role was to get out of his way and help hone his innovation and convey his work to the outside world.”
Together, Kartin and Conway were able to bring this new framework to life, dubbing it optimistic machine learning (opML). Once implemented, the opML approach to on-chain machine learning immediately showed a significant improvement over zkML. They couldn’t believe the contrast in hardware requirements for their model versus zkML – opML could run a large language model (e.g. 7B-LLaMA) on a common PC without a GPU. In contrast, zkML would take nearly 80 minutes to generate a proof for a much smaller, 1M-nanoGPT model on the same hardware.
Given this computational efficiency, opML has been able to unlock novel use cases for on chain machine learning like content authenticity, on chain AI games, sensitive data processing (e.g. healthcare), and future use cases they’ve only just scratched the surface of.
Reflecting on my conversation with Kartin, I can’t help but wonder how he was able to so adeptly work with Conway to coax this groundbreaking innovation out of him. How was he able to translate this vision when the world’s leading venture fund couldn’t even make it past one article? I think it has something to do with how Kartin’s myriad identities allow him to connect with people from all walks of life. His technical background allowed him to understand the potential in an overlooked forum post while his soft skills meant that he was also able to take that diamond of an idea and polish it into opML. When I ask him about where he thinks these abilities come from, he takes me back to his childhood.
“Growing up I thought I was going to be a musician. I actually had a contract from a recording label around the time of the Taylor Swift performance. So, I think it’s just in me to perform. I have to perform. And when I perform, I have to figure out my audience and how to connect with them.”
Reflecting on what he’s most excited about with ORA’s future, Kartin tells me there is much more to come with opML. “It’s kind of new. It’s like a newborn baby," he explains. “It just came out this past April so we need to put many more resources into building out the opML library. There are bottlenecks that we just resolved but it’s going to be a bit before we can announce them.”
By the end of my time with Kartin, I can’t help but share his excitement. Having already built and refined their zkOracle platform, his team has rapidly pushed forward into uncharted territory with opML. This product has the potential to radically transform the visibility and verifiability of artificial intelligence. It’s a big problem, one that necessitates a strong founder who can set a vision in a dynamically shifting landscape. I couldn’t think of anyone better equipped to tackle that challenge than Kartin. ✦