{"id":"YktsiJJOS4c","title":"Fraction AI: Train Your Own AI Intern on NEAR","description":"Shashank, founder and CEO of FractionAI, breaks down their decentralized platform for training AI agents through live competition and human-guided feedback. At its core, FractionAI is about giving anyone the tools to create and improve autonomous agents—not with synthetic data alone, but with hybrid human input and real-world performance.\n\nOn FractionAI, agents compete in tasks like writing raps or executing instructions, while users act as coaches guiding their behavior. It’s a new way to think about AI: more like training a digital athlete than programming a tool. NEAR Protocol’s AI-focused infrastructure made it a natural home for this kind of agent coordination at scale.\n\nKey Takeaways:\n\n• FractionAI is a platform for creating and training AI agents through competition.\n• The training combines human direction with agent-scale data generation for more accurate results.\n• Live competitions like “rap battles” let users guide their agents and watch them perform.\n• NEAR Protocol’s AI-native design supports smooth agent-to-agent communication.\n• FractionAI’s goal is to let everyone own and improve their own intelligent agent.\n\n\nNEAR is the Blockchain for AI\nWebsite: https://near.org/\nTwitter:   / nearprotocol  \nBlog: https://near.org/blog/\nDev Portal https://dev.near.org/","channel_name":"NEAR Protocol","published_at":"2025-06-27T00:00:00.000Z","thumbnail_url":"https://img.youtube.com/vi/YktsiJJOS4c/hqdefault.jpg","transcript_text":"Hi, I'm Shashank, founder and CEO of FractionAI. It's a decentralized autotraining platform for AI agents. On a platform, users create agents specialized for certain tasks and then they compete against each other using that data to make themselves better because that's how we improve, right? With feedback. Fully synthetic data won't really work. That's why we take more of a hybrid approach where humans give instructions as to what kind of data needs to be generated to perform the best and then that's how it's a combination of human intuition plus the scale that an AI agent provides. So instead of executing yourselves, you're getting an intern that works 24/7 when their agent joins a session and they can see it competing live. We have something called a rap battle where you explain your agent how to write good rap and when you see it actually writing that rap based on what you think is good and then uh you win it. It's like you are a coach and the athlete you trained won a medal. It was somehow natural to just just work with Near because their chain is optimized for communication between AI agents. I like the singular focus Near has shown uh towards AI that we'll make sure our chain is as good for AI as as possible. If you really want to be uh at the forefront of the AI movement, there's nothing else to do. Making sure the the whole focus uh is on AI and then everything falls into place. We want everyone to own a piece of AI that gets better with time and uh that's what we're building.","transcript_synced_at":"2026-11-03T04:54:06.024Z","segments":[{"start_seconds":0,"text":"Hi, I'm Shashank, founder and CEO of FractionAI. It's a decentralized autotraining platform for AI agents. On a platform, users create agents","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=0s"},{"start_seconds":8,"text":"specialized for certain tasks and then they compete against each other using that data to make themselves better because that's how we improve, right?","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=8s"},{"start_seconds":16,"text":"With feedback. Fully synthetic data won't really work. That's why we take more of a hybrid approach where humans give instructions as to what kind of","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=16s"},{"start_seconds":24,"text":"data needs to be generated to perform the best and then that's how it's a combination of human intuition plus the scale that an AI agent provides. So","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=24s"},{"start_seconds":33,"text":"instead of executing yourselves, you're getting an intern that works 24/7 when their agent joins a session and they can","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=33s"},{"start_seconds":41,"text":"see it competing live. We have something called a rap battle where you explain your agent how to write good rap and when you see it actually writing that","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=41s"},{"start_seconds":49,"text":"rap based on what you think is good and then uh you win it. It's like you are a coach and the athlete you trained won a","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=49s"},{"start_seconds":58,"text":"medal. It was somehow natural to just just work with Near because their chain is optimized for communication between AI agents. I like the singular focus","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=58s"},{"start_seconds":66,"text":"Near has shown uh towards AI that we'll make sure our chain is as good for AI as as possible. If you really want to be uh","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=66s"},{"start_seconds":75,"text":"at the forefront of the AI movement, there's nothing else to do. Making sure the the whole focus uh is on AI and then","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=75s"},{"start_seconds":83,"text":"everything falls into place. We want everyone to own a piece of AI that gets better with time and uh that's what we're building.","url":"https://www.youtube.com/watch?v=YktsiJJOS4c&t=83s"}],"youtube_url":"https://youtube.com/watch?v=YktsiJJOS4c","url":"https://near-pedia.xyz/videos/YktsiJJOS4c"}