The models that will plow fields, repair infrastructure, and navigate Mars won't be trained on the internet. They need real-world, event-based data — collected and structured for machines that must act with limited compute and limited information. That's what Phasor & Episodes are.
High-quality training data for large language models is running out. The models that dominated the last decade were built on assumptions — infinite data, cheap compute — that are collapsing. The AI systems being built now need the right real-world data — and that data has, until now, barely existed.
Episodes is a collection of real-world, event-based datasets designed to train Lean AI — models that learn to act from sparse, structured signals rather than massive text corpora.
Whether you're training lean AI models in Academia or privately, we have the data you need. We know the pain of trying to find and prepare your own dataset. Leave that to us — Episodes gives you something to start with. And Phasor gives you the pipeline to go further.
Explore the dataset →We know that if you're doing research right now, you need data. We literally want you to tell us what data you need and we will get it to you.
Email us your name, institution, and request to v eric at projectphasor dot com or fill up this form
Phagle is Phasor's answer to a Neuromorphic AI open competition — built for neuromorphic and data-efficient AI developers. Episodes is the official benchmark dataset. If you're a developer, researcher, or ML engineer working at this frontier — this is your competition.
We needed what Phasor is building ourselves. After a year of exhaustive research and 130+ interviews, we didn't find an alternative — so we built it.
Whether you're a researcher, developer, or just curious about what's next for AI — there's a seat at the table. We're building in the open and we'd love to have you along for the ride.