We are a multi-modal human interaction data research lab.

Our mission is to bring AI into the real world by capturing and modeling multi-signal human activity and decision-making processes → combining physical motion, physiological signals, gaze/vision, speech, intuition and contextual cues.

AI is meaningful when it can be a real teammate, fully understanding what people are doing, how they decide, and how they interact.

PROCESS

We capture and structure the human layer of intelligence

Define
Focus on critical aspects of human activity, attention, and decision-making that today's AI can't yet model.
arrow
17 proprietary dataset
100+ petabytes
500M+ participants
arrow
17 proprietary dataset
100+ petabytes
500M+ participants

Our datasets are used by leading AI labs and research teams developing models in multimodal learning, cognitive modeling, human-AI collaboration, and decision-making analysis, enabling AI that becomes a real teammate.

Explore our datasets

Multi-modal Conversational and Human interaction Data

Billions of sequential messages in 100+ languages from 500+ million users, augmented by views, forwards, replies, reactions and media files.

Great for: model cooperation with users, multimodal alignment, conversational grounding

Diverse Video Data with Meta Information

9,000 years of video from 300M+ clips. Categories contain computer use, gaming, spokesperson, DIY, negotiations, and many more.

Great for: computer/browser use, reasoning, video generation, lip-sync, etc.

Sport events: well structured and labeled

Thousands of hours of football, soccer and other sports in high quality and precise labeling.

Great for: coordination and strategy modeling, sport actions video generation, multimodal human interaction alignment across languages

Medical images / Lab test results / Medical records

45M+ studies from 10M patients showing medical history of patients over time

Great for: Medical AGI, AI Medical Assistants