Egocentric data to train the next generation of physical AI.
We're a data operation for physical AI. We record egocentric, first-person video of real tasks in real environments, and turn it into training data for robots and embodied AI. Contract-based, shot to spec, priced per hour of usable footage.
Physical AI and robotics are moving fast, but progress is bottlenecked less by weak models than by a lack of high-quality, real-world training data. Good robots need egocentric, first-person data of real tasks in real environments, and almost nobody can collect it at scale, affordably, and to spec.
Today, it's focused capture: a client specifies the task, the environment, and the format, and we go get it. Later, it becomes a standing library: any process recreated and generated on request, without a fresh shoot every time.
A few of the environments we shoot in:
Prior experience collecting data to train voice and video models. I know what good training data looks like and how hard it is to source.
At Build AI, our team collected over 1M hours of egocentric data. I personally contributed 500K+ hours of that.
Beyond that, a network of business and community relationships gets us into the exact locations and environments clients need.
It was never really about the camera. It's access to the people, places, and processes worth recording, and that only comes from relationships built over time, not from a product.
Abhinay Tiwari, founder. 100% ownership, solo, full-time. Writes all the code and does all the technical work personally.
Open to raising, and open to collaborating with teams building physical AI, robotics, or embodied AI data pipelines. Happy to share a longer deck or sample data on request.
Book a 30 minute call or connect on LinkedIn.