This is exactly right.
Collecting high-quality narrated egocentric data from real skilled trades work inside occupied homes is extremely operationally heavy. It requires licensed crews, consistent job site access, reliable upload pipelines, and daily quality control. It’s not a side project or marketing stunt.
Most people dramatically underestimate how hard the “manual ops” part actually is at scale.
Gearing up for another long weekend of data collection.
This is the real world footage robots will actually need to do skilled trades work HVAC, plumbing, and electrical inside real commerical job site today. Not simulations. Real decisions, real environments, real narration from licensed techs.
Building the data foundation for physical AI in blue-collar work.
@sridharfyi Solo founder here. I’m building Tradeye — we capture narrated first-person video from licensed HVAC, plumbing, and electrical technicians working inside real occupied homes and commercial buildings. High-quality, legally compliant data for humanoid robots and embodied AI.
https://t.co/trLmSdHEA3
The biggest technical decision was using Ray-Ban Meta smart glasses for egocentric capture combined with real-time narration by licensed trades technicians.
Most robotics datasets are either silent third-person video or simulated. By having skilled HVAC, plumbing, and electrical techs naturally narrate their actions, tools, and decision-making while working in real occupied homes, we capture intent and reasoning — not just motion. That’s the signal humanoid robots actually need.
The California C20 license also lets us film legally on active job sites, which is a major moat most teams don’t have.
Fully aligned with this direction.
We’re building a dataset of narrated, first-person video from real HVAC, plumbing, and electrical work captured on actual job sites under a California C20 contractor license.
This gives high-quality, long-horizon physical data that’s very hard to generate through simulation alone.
https://t.co/3AyBs7lXHn
@HartmansDoeke Solo founder building a dataset of narrated, first-person video from real HVAC, plumbing, and electrical work captured on actual job sites under a California C20 contractor license.
Creating high-quality training data for humanoid robots and embodied AI.
https://t.co/LtDJOQ7jaP
The shift toward real-world understanding is exactly why we started Tradeye. We capture narrated first-person video from actual HVAC, plumbing, and electrical work in occupied homes and buildings — all under a California C20 contractor license.
This gives high-quality, long-horizon physical data that simulation alone can’t produce.
https://t.co/3AyBs7lXHn
#EmbodiedAI #PhysicalAI #HumanoidRobots #WorldModels
Congrats on the $400M raise — huge milestone.
We’re building exactly the kind of real-world data that companies training physical AI need. Tradeye captures narrated first-person video from real HVAC, plumbing, and electrical work in occupied homes and buildings under a California C20 contractor license.
This gives high-quality, long-horizon trades data that simulation can’t replicate.
Would love to share what we’re building if it’s relevant.
https://t.co/3AyBs7lXHn
@NVIDIARobotics
Simulation is getting faster, but real homes are still messy.
At Tradeye we’re building the real-world data layer these models need: https://t.co/3AyBs7lXHn
We capture narrated first-person POV footage from licensed HVAC, plumbing, and electrical crews on actual occupied job sites under our California C20 contractor license. This creates the authentic, long-horizon, decision-heavy data that complements simulation and helps close the sim-to-real gap.
Happy to explore how our trades dataset could support NVIDIA’s physical AI work.
@shenhaichen@gi_labs
Nice work on Ego1.
We’re building the real-world trades counterpart at Tradeye: https://t.co/3AyBs7lXHn
We capture narrated first-person POV footage from licensed HVAC, plumbing, and electrical crews on actual occupied job sites under our California C20 contractor license. This creates messy, long-horizon, decision-heavy data that complements hardware-focused egocentric capture systems.
Happy to explore how our dataset could help ground models trained on tools like Ego1.
@compileandpush
Honestly? The boring part that almost killed it was consistent data organization and labeling.
Early on I was capturing footage but falling behind on labeling and structuring it properly. It started piling up and became overwhelming. I had to force myself to treat the “boring” work (daily labeling, folder structure, metadata, master tracking sheet) as non-negotiable — same as showing up to job sites.
That shift is what’s letting us actually release usable datasets now.
Appreciate you calling it out.
Still grinding every day.
I’m still out on job sites, still running crews, and still trying to save the family business the best way I know how.
Tradeye isn’t just a side project for me — it’s my shot at building something that can actually help keep skilled trades alive and make the work more sustainable long-term. The data we’re capturing from real homes and real jobs is the foundation for that.
Some days it feels slow. Most days it’s just work. But I’m still here putting in the hours.
Appreciate everyone who’s been following along.
https://t.co/3AyBs7lXHn
@XSquareRobot@QianWangX2robot
Nice work on WALL-WM.
We’re building the real-world data layer these models need at Tradeye: https://t.co/3AyBs7lpRP
We capture narrated first-person POV footage from licensed HVAC, plumbing, and electrical crews on actual occupied job sites under our California C20 contractor license. This creates messy, long-horizon, decision-heavy data that complements simulation and lab-based world models.
Happy to explore how our trades dataset could help ground models like WALL-WM.
@MartinGTobias@MartinGTobias Building Tradeye to save my family’s HVAC business. We capture real narrated work from licensed crews on actual job sites so robots can learn from the messy real world. The revenue goes back into keeping the business alive. https://t.co/LtDJOQ7jaP
@patmatthews Just applied to Active Ventures. We capture narrated egocentric data from licensed skilled trades crews working inside real occupied homes and commercial buildings. Applied to Speedrun but didn’t get in. Building this to save and grow our family-owned HVAC business while creating the real-world dataset humanoid robots need. https://t.co/3AyBs7lpRP