The startup filming people's chores just raised $47 million

5 Sources

Share

Humyn Labs, a months-old startup that records people doing everyday work to train robots, has raised $47 million in a round led by Accel. The split is $26 million of Series A equity and $21 million of venture debt, the debt led by Lighthouse Canton Growth Debt Fund II with Trifecta Capital taking part

1

5

.

Source: ET

Source: ET

Why raise at all? "This business has very positive gross margins and generates EBITDA, so we were looking to scale it through our own cash flows. But the business has scaled beyond our expectations," co-founder and CEO Manish Agarwal told Economic Times

3

. The company says it crossed $50 million in annualized revenue within six months of operations

1

3

. The more concrete figure, per Agarwal: $10 million in actual revenue between April and October, plus another $20 million in signed contracts

1

. He credits the pace to what he calls hustle and working with more frontier labs, in a market he sizes at $10 billion a year

3

.

How the data gets made

Agarwal, previously chief executive of Indian gaming company Nazara Technologies, founded the company in early 2026 with Ishank Gupta, a former executive at brewer Anheuser-Busch InBev

1

3

. Kratos Studios, which later became KGeN, is the pair's earlier venture, and the founders credit its contributor network for helping collect data in people's homes

1

.

The company built its own two-camera, head-mounted rig to film people at work, then uses software to label the footage and track hand movements and object interactions. Collection runs in more than 20 countries, with Brazil, India and South Korea as the main hubs

1

2

. Its datasets include up to 200,000 hours of captured human motion, plus a 120,000-face dataset for facial recognition. There are also more than 50,000 hours of speech in underserved languages such as Bengali, Tamil and Korean, per SiliconANGLE

2

.

Source: SiliconANGLE

Source: SiliconANGLE

The buyers are mostly American. About 90% of revenue comes from the US, per Gupta, from 12 paying customers that include frontier AI labs and robotics companies; none are named, citing NDAs

1

. Other startups, among them Mecka AI and MicroAGI, are chasing the same data, and India has emerged in recent months as a hub for this kind of collection work

1

.

From data seller to deployment partner

"We started eight to ten months ago with data collection, but we are clear that data collection itself is not the real game. The real game is building a research lab, establishing deployment centres and controlling input quality at the point of collection," Agarwal said

3

.

The planned spending covers deployment centres in Brazil, South Korea and India, further development of the capture hardware, and new teams in San Francisco, London and Seoul

3

2

4

. The idea is to test where robots fail in the wild and feed the results back into the models

2

. Agarwal argues that a robot trained in one country's homes may fail in another's kitchens and laundry routines, and that mistakes in the physical world can be catastrophic where chatbot mistakes are not

2

. A July Counterpoint Research estimate says India's physical-AI data services opportunity could be worth up to $10 billion by 2030

3

4

. Agarwal says the company aims to build a $1 billion annual-revenue business but gave no timeline

3

.

The debt side exists, Agarwal says, because the company pays workers and collection partners daily or weekly but waits 30 to 60 days for payment from its AI lab customers

1

.

The people on camera

Recording people, sometimes inside their homes, has raised questions about privacy, consent and who owns the resulting data

1

4

. Inc42 points at the contradiction in the arrangement: the workers whose motion is recorded are helping train systems that could one day automate their own jobs in warehousing, packaging, assembly and cleaning

4

.

Agarwal says the company films people doing their usual work or chores rather than staged activities, using local partners to find sites and supervise collection

1

. "The worker is a worker who is incentivized, but he needs to be monitored and supervised for the right input," Agarwal said

1

.

Source: TechCrunch

Source: TechCrunch

© 2026 TheOutpost.AI All rights reserved