Real work.
Real skills.
AI learns from both.
Eyes of Work is an early-stage applied AI research venture in Mexico building a human-centered network for collecting high-quality, rights-cleared demonstrations of real physical work.
Turn human skill into a resource for the AI economy.
People who clean, build, repair, maintain and move the physical world hold practical knowledge that is rarely represented in AI training data. We are exploring a model where skilled people can earn additional income by contributing carefully consented demonstrations of their work — while AI teams gain access to more authentic physical-world data.
For skilled contributors
A new way for independent workers to benefit from the value of their experience today, as robotics and automation increasingly reshape physical work.
For Physical AI teams
Access to diverse, real-world demonstrations that include natural variation, recovery from mistakes, tool use, decision points and changing environments.
Start with people who already know the work.
Our initial network in Quintana Roo is intentionally small and hands-on. We are validating collection with independent professionals before scaling.
Automotive repair
Diagnostics, tools, disassembly, reassembly and verification.
4 mechanics accessibleHVAC service
Inspection, maintenance, measurement and repair workflows.
Field accessConstruction
Measuring, drilling, fastening, materials and two-handed tool use.
Independent workersProfessional cleaning
Folding, sorting, object handling, sequence completion and recovery.
Independent workersThe goal is not “more video.” The goal is structured evidence of how skilled humans perceive, decide and act in the physical world.
Define
Start from a robotics or research need, not speculative bulk collection.
Capture
Record bounded, consented physical tasks with the sensor package required.
Structure
Synchronize signals, segment actions, map objects, outcomes and context.
Deliver
Provide QA-approved research data under clearly defined usage rights.
A field research layer for embodied intelligence.
We are investigating whether decentralized collection from real skilled workers can become a practical source of multimodal demonstrations for action understanding, vision-language-action systems, imitation learning and robotic skill acquisition.
Initial research direction
We are evaluating research-grade wearable sensing — including Project Aria — alongside lower-cost open capture rigs. The objective is to understand which signals materially improve downstream task understanding and robot-learning usefulness, and which can be collected economically at scale.
Consent and rights are part of the dataset.
We do not treat privacy as a post-processing step. Our intended model requires informed contributor consent, permission for the recording environment, bounded collection tasks, privacy controls and explicit downstream usage rights.
No hidden collection
Participants know what is recorded, why it is recorded, how it may be used and how they are compensated.
Rights-cleared delivery
Datasets may be licensed to one or multiple research customers only where contributor and environment agreements explicitly permit that use.
Build the first pilot with us.
We are looking for Physical AI and robotics teams that can define the sensor configuration, tasks and metadata most useful to their models. We can then run a small controlled collection in Quintana Roo before scaling.