What it would take to become India's largest human data provider

Zirconoid intends to become India's largest human data provider. That line is easy to write and hard to mean. Open egocentric hubs already fight over published hour columns, and they will keep doing so. For us, largest means the densest supply of rights-cleared, domain-matched, human-captured data from the places where India actually works: small factories, large industries, and households. Access to work, with a license a buyer can train on.

Physical AI still learns badly from environments it never enters. Labs can rent headsets and teleop rigs this quarter. Standing up consenting operators inside a running plant, a contract workshop, or a domestic kitchen takes longer, and that delay is where most capture programs stall. Cameras arrive in days. Floors take relationships, consent, and a schedule that fits production. Anyone who has tried to film an SOP on a live line already knows this.

India makes the access problem unusually large. The inventory is real. The latest published Annual Survey of Industries is still ASI 2023-24 (released August 2025): roughly 260k registered factories and about 19.6 million people engaged. ASI 2024-25 results are not out yet, so we do not invent a newer factory count. Under that frame sits a thicker layer. QBUSE for January through March 2026 puts unincorporated non-agricultural establishments at about 91.6 million, with about 151.7 million workers. PLFS 2025: roughly 616 million employed ages 15 and up, 56.2% self-employed, manufacturing at 12.1% of employment. Udyam as of August 2026: about 91.6 million MSMEs and 400 million-plus in employment, with MSMEs around 31% of GDP and about 35% of manufacturing output (Economic Survey 2025-26).

That is why India is a capture geography. It does not tell you who answers the phone at six in the morning.

Cold outreach does not open a running plant. It does not open a household kitchen for scoped egocentric capture either. Consent, schedule fit, and a manager who will clear a live shift take trust. Our India access sits on personal and professional networks built through lineage, family, and long professional engagement over roughly the last forty years. Those relationships are now turning into capture relationships: introductions that clear plant management, operators who will wear a camera on a live shift, and households that treat filming as a job with a brief rather than ambient scrap. A mid-layer data company or robotics lab buying India coverage this quarter needs floors they can train on. We recruit operators and partner with the sites where the work already runs - workshops, factories, plants, bottling lines, mills, logistics hubs, kitchens, and homes - because the work that matters happens inside buildings other people own.

We run one method across site types. Small factories and workshops come first in volume and variance. A four-person electronics bench and a twenty-person injection-mold cell teach different habits for the same task: material changes, tooling changes, recoveries from mistakes. We capture around the production schedule rather than across it. Large industries extend the same method into plants that already have security, shift discipline, and long SOPs. A signed site gives a roster, a schedule, a floor we are allowed to film, and a manager on the morning call. Households close the continuum. Kitchens, home workshops, and residential routines are where many embodied models will eventually live. India's self-employed majority puts domestic work in the same recruiting problem as the factory floor. We treat homes as sites. Consent, rights, and a task brief still come first.

Rights, coverage, labels, and delivery are how density becomes a training pack. Every pack clears operators and sites before it ships - a buyer training commercial weights should not chase re-consent after the fact. Coverage means domain match across the continuum so a task family is not locked to one site's habits; breadth of site changes what the model sees, not only how fast footage accumulates. Labels means clean video plus separate sidecar labels in JSON or CSV. Burned-in overlays look useful in a demo reel and fight the training pipeline. Delivery means scoped samples, capture plans, and packs a mid-layer data company or robotics team can put into a program this quarter.

Open hubs will keep publishing larger hour columns. That is a useful research signal. Hours from hubs and hours from running production floors remain different goods for buyers who ship industrial or domestic SOP packages. We are organizing the second kind: human-captured egocentric data at ops and geo scale, with India as the densest near-term inventory and global regions as we go. Early samples from real factory and workshop work are at zirconoid.com/samples.

Engineering systems that solve technical problems get commoditized. Fresh operator-collected data does not, because it is specific to the person, the place, and the task. Becoming India's largest human data provider, on that definition, is an access problem first. India supplies the density. Decades of trusted relationships supply the way in. Rights, coverage, labels, and delivery decide whether density becomes usable training data.

If the domain fits a program you are scoping, write to [email protected].

Sources.

ASI 2023-24 (MoSPI / PIB, August 2025): about 260k factories and about 19.6 million persons engaged. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2161192

QBUSE January–March 2026 (NSO / MoSPI, May 2026): about 91.6 million unincorporated non-agricultural establishments and about 151.7 million workers. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2263720

PLFS Annual Report 2025 (NSO / MoSPI, March 2026): about 616 million employed ages 15 and up, 56.2% self-employed, manufacturing 12.1% of employment. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2246009

Udyam and Economic Survey 2025-26 (MSME / PIB, August 2026): about 91.6 million MSMEs, 400 million-plus in employment, about 31% of GDP and about 35% of manufacturing output. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2296358 and https://www.pib.gov.in/FactsheetDetails.aspx?Id=150826&lang=8&reg=20

ASI 2024-25 fieldwork ran October 2025 through May 2026; results are not yet published.