Founded in February 2026 by former Nazara Technologies CEO Manish Agarwal and former AB InBev executive Ishank Gupta, Humyn collects recordings of people performing everyday activities to help AI systems learn how humans interact with physical environments. The founders previously established Kratos Studios, now called KGeN, whose contributor network helps Humyn recruit participants across emerging markets.
The company's approach focuses on capturing natural human behavior rather than relying exclusively on controlled demonstrations. Participants wear camera equipment while completing ordinary household or workplace activities, allowing Humyn to record how people handle objects, coordinate movements and respond to their surroundings.
Humyn has developed its own two-camera recording hardware, supported by software that processes footage and identifies details such as hand movements and object interactions. The company says its collection of first-person video includes as many as 200,000 hours of human activity.
The resulting information is designed to help train physical AI systems, including humanoid robots, that must understand more than visual recognition. These systems need to learn how objects are manipulated, how much pressure different actions require and how movements change depending on environmental conditions.
Agarwal emphasized that errors involving physical machines can have more serious consequences than mistakes made by conversational AI systems.
“The results can be very, very, very dramatic and catastrophic,” Agarwal told SiliconANGLE.
Humyn's data platform combines several types of information, including video, audio, movement measurements and grip-force readings. Its processing system validates, labels and synchronizes those inputs to create datasets that developers can use for robotics, world models and other AI applications.
The company supports multiple data formats, including MCAP, RLDS, LeRobot v3, ROS Bag, WAV, FLAC and JSON, allowing customers to incorporate the information into different development environments.
Geographic diversity is another part of Humyn's strategy. The company operates collection programs in more than 20 countries across Asia and Latin America, with Brazil, India and South Korea serving as its primary centers.
Local partners recruit participants, identify suitable recording locations and oversee collection activities. Humyn argues that capturing everyday behavior across different regions can help robots learn to operate in environments with varying household layouts, equipment and cultural practices.
For example, a robot trained primarily on activities recorded in American homes may encounter unfamiliar cooking tools, appliances or cleaning routines when deployed in India. Collecting examples from different settings gives developers additional information about those differences.
“Diversity gets you in the room. The quality ensures that you stay in the room,” Agarwal said.
Beyond robotics, Humyn has assembled datasets for facial recognition and speech AI. The company reported delivering recordings representing 120,000 faces across five regions, incorporating variations in lighting, appearance and other visual characteristics.
It has also collected more than 50,000 hours of speech data covering languages including Bengali, Tamil, Telugu, Odia, Assamese, Korean, Tagalog, Arabic, Vietnamese and Portuguese. These datasets are intended to support voice applications in markets where existing training resources are less extensive.
Humyn says demand is coming from developers building household robots, elder-care systems and industrial machines designed to handle objects or navigate workplaces such as factories and warehouses.
According to Gupta, the company currently serves 12 paying customers, including advanced AI research labs, robotics developers and companies building world models. Approximately 90% of its revenue comes from the United States, although Humyn has not disclosed its customers' identities.
Agarwal said the startup generated $10 million in revenue between April and October 2026. It also has $20 million in signed customer contracts and another $20 million in potential agreements that have not been finalized.
That growth has created financing requirements for the company's collection operations. Humyn typically compensates workers and local partners daily or weekly, while customers generally take between 30 and 60 days to settle their invoices.
Agarwal explained that the venture debt portion of the financing will help cover the resulting cash-flow gap without requiring the company to raise the entire amount through equity.
Humyn currently employs 51 full-time staff members, most of whom work in Bengaluru, and relies on additional local partners throughout its international collection network.
The company intends to expand beyond collecting and preparing training data by developing systems that evaluate how robots perform after deployment. Its plans include using simulations, physical testing environments and collaborations with researchers and businesses to identify differences between expected robotic behavior and actual results.
“The next phase in the deployment cycle is about how well these models perform in the wild,” Agarwal said.
Humyn plans to use the funding to strengthen its hardware and data-enrichment systems, build testing infrastructure in its three primary collection markets and expand its presence in San Francisco, London and Seoul.
The investment gives Humyn additional resources to support its growing customer base while addressing the costs of operating a distributed data-collection network. Its next stage will focus on connecting human activity recordings with real-world robot testing, allowing developers to use observed performance problems to improve their AI models.
This analysis is based on reporting from Silicon Angle.
Image courtesy of Humyn Labs.
This article was generated with AI assistance and reviewed for accuracy and quality.