DeepReach Inc.

Alumni Q&A

How did your company get started? (i.e., How did the founders meet? How did you come up with the idea? How did you decide to be a founder?)

The two founders are from the same hometown and have known each other for ten years. One previously built HireIO, a cross-border staffing platform, and Talex, an LLM data-annotation expert network of roughly 70,000 people across ten countries; the other holds a CS PhD in computer vision from USC and worked on ML infrastructure at Meta before becoming a founding ML engineer at Aven. Working on the supply and model sides respectively, both concluded that robot training needed data captured from many different real workplaces rather than a single lab, and they started DeepReach in 2025, splitting responsibilities between perception/data and network/hardware/commercial.

What is the core problem you are solving? Why is this a big problem? What made you decide to work on it?

The founders address the problem that robot models fail to generalize because their training data comes mostly from single, controlled environments, making diversity rather than volume the binding constraint. They note the field has roughly tens of thousands of hours of physical-work data against an estimated need in the hundreds of millions, and that customer acceptance specs now cap how much of a delivery can come from one environment type. Their approach builds a network in which people already working in varied workplaces collect data locally as their own business, using hardware, software, and payments the company provides; the founder came to it after building cross-border human networks at two prior companies.

What is your long-term vision? If you truly succeed, what will be different about the world?

The founders aim to build a network reaching workplaces worldwide, targeting more than 10,000 devices across 50 countries, one million commercial scenes, and ten million task types, and to compile a record of how physical work is done, called the Human Skill Library. They intend for robot policies trained on many real environments to work in ordinary small businesses rather than only large facilities built around machines, and for the workers who supply the data to earn from it; they report data partners in ten countries already operating this as their own business.