From Nepal to the Bay: the engineer teaching robots what’s worth watching

Divya Pandey left systems engineering to co-found O’WOW, a company betting the future of Physical AI turns on the right data, not more of it. Robots don’t fail for lack of data. They fail because the data they learned from never put them in the situations that matter. A machine can watch a thousand hours of footage and still fumble the one moment that counts: the knot that won’t tie, the corner it never saw in training. The scarce resource was never volume. It’s knowing which experience would change how a machine behaves, and proving that it did. That problem is where Divya Pandey works. The Nepal-born engineer is co-founder and COO of O’WOW, a Palo Alto company building Data Intelligence for Physical AI. An engineer trained to spot what breaks Divya came to the United States for her master’s and started as a systems engineer, building computer systems that could not go down. The work taught her to see the world as systems: inputs, failures, feedback, repair. When she looked at how AI companies sourced training data, she saw a system broken in a specific way. Everyone was racing to collect more. Almost no one could say which of it mattered. That instinct became O’WOW’s method. Today she runs the engine room: a capture-and-verification network across more than 15 countries, the pipelines that check each recording in layers, and the people who do the work, each country with its own rules on jobs, taxes and payment. Research and operations run as one function, and she owns it. Teaching robots the things they can’t read O’WOW is a research-led data company for Physical AI, the robots and machines that act in the real world rather than answer in a chat window. It starts from where a model fails and runs a loop: define the task, investigate the failure, collect against that exact gap with traceable provenance, validate in layers, then measure the next model version, and repeat. A clean recording isn’t the finish line. Data can be technically perfect and still miss the task, the environment, or the failure that mattered. Divya’s job is to hold the whole chain together, from capture on a street in Kathmandu to the proof that it was worth capturing. The partner from Ballia She didn’t build it alone. Her co-founder and CEO, Gangesh Pathak, a self-taught entrepreneur from Ballia, India, built and sold his first company before turning to the data behind AI. Where Divya builds and runs the operation, Gangesh sets product direction, AI strategy and growth, and brought in the first customers. Two temperaments, one company. A door that opens from home For Divya, the mission runs past one company. The same network that serves frontier labs opens doors from places old hiring systems never reached. “AI will keep depending on people,” she says. “Nepal shouldn’t only use AI. Our young people can help build it, from home.” The post From Nepal to the Bay: the engineer teaching robots what’s worth watching appeared first on Technology Khabar.