India’s Human Archive Trains Robots; FirstCry Cuts Losses

Indian Startups: Human Archive Pioneers Robot Training & FirstCry Boosts Financials

The Indian startup ecosystem is witnessing significant developments this week, with Human Archive gaining prominence for its innovative approach to training global robotic and physical AI systems using data collected from Indian workers. Simultaneously, FirstCry, a prominent e-commerce platform, has reported a substantial financial turnaround, narrowing its losses by 61% in the fourth quarter of FY26, alongside a robust 12% increase in revenue. These distinct narratives highlight both the cutting-edge technological advancements originating from India and the improving financial resilience of its digital enterprises.

Human Archive’s Data-Driven Approach to Physical AI Development

Human Archive, a San Francisco and Bengaluru-based startup backed by Y Combinator, has secured $8.2 million in a funding round led by Wing Venture Capital and NVP Capital. This investment is earmarked for developing what the company describes as the largest human sensorimotor dataset of its kind. Founded by UC Berkeley and Stanford University dropouts Rushil Agarwal, Raj Patel, Samay Maini, and Shloke Patel, the startup’s model involves recording Indian workers performing various home services tasks to generate training material for advanced robots and physical AI systems that frontier labs worldwide are rapidly developing.

The company’s operations have recently come under public scrutiny following reports of home services startups, such as Pronto and Snabbit, conducting pilots using worker tracking data for AI training. While Pronto reportedly ran a pilot using worker data, Snabbit, in partnership with Human Archive, conducted similar tests in “controlled environments,” with the human services startup denying real-time deployment. Human Archive’s business model is centered on collecting vast amounts of data, aiming for millions of hours beyond the tens of thousands already amassed, to capture “human embodied intelligence,” as co-founder Rushil Agarwal stated.

Ethical Considerations and Data Collection Methodology

Human Archive employs a rigorous data collection process, supplying workers with specialized hardware rigs that include downward-facing 4K cameras, depth-sensing cameras, and wide-angle lenses to record human hands performing specific tasks. This data, which captures patterns and movements, is then used to train robots and physical AI systems. The company emphasizes that faces are rarely visible due to camera angles, and any identifiable information is stripped out during post-processing to address privacy concerns. Additional hardware components such as tactile gloves, wrist-mounted cameras, and body IMUs further enhance the dataset’s richness.

Once footage is recorded, it undergoes extensive processing, including motion capture technology and tactile force-feedback streams, to produce a high-fidelity dataset. This data is further refined in-house by Human Archive through proprietary QA, hand-tracking, and reconstruction models. India serves as the largest hub for Human Archive’s data collection efforts due to its diverse industries, ranging from jewelry and textile to coal and steel, and a robust labor economy that is unique compared to regions like the US. The startup has partnered with over 125 companies, primarily in India, across sectors like hotels, restaurants, quick commerce platforms, construction sites, and factories, although many partnerships are yet to be activated.

Worker Compensation and Privacy Dilemmas in India’s Blue-Collar Sector

The ethical implications of workers unknowingly training machines that could eventually displace them have become a significant point of debate. Human Archive maintains that workers directly engaged by the company are fully informed about data capture. However, when data collection occurs through partner businesses, the responsibility for worker education shifts. Rushil Agarwal stated that while they are transparent with partners regarding data usage intent, it is the partner companies’ responsibility to educate their employees. Contracts reportedly require partners to obtain informed consent, and Human Archive claims to work only with compliant entities.

Regarding compensation, Agarwal mentioned that workers receive an average of $10-100 per hour, though this could not be independently verified. This compensation is noteworthy given that blue-collar workers in India typically earn between ₹15,000 to ₹35,000 per month, with Deloitte’s 2025 trends report indicating an annual wage growth of 5–6%. The privacy challenge is further exacerbated when recording occurs in private residences, as seen with housekeeping services, potentially conflicting with regulations like the DPDPA 2023. The Ministry of Electronics and Information Technology (MeitY) is reportedly scrutinizing the Pronto controversy, which could lead to increased oversight for startups utilizing customer home data.

The Growing Landscape of Physical AI Data Collection

The race to develop physical AI systems has created a new category focused on data collection. Unlike language models trained on internet-scraped tokens or vision models learning from public images, robotic manipulation data necessitates real-world capture. Companies like Scale AI, with a 49% Meta stake, operate a dedicated Data Engine for Physical AI, having completed over 100,000 production hours at its San Francisco facility. Build AI, a newer entrant, offers egocentric datasets of first-person video tailored for industrial robots and embodied AI.

The physical AI market is projected for substantial growth, with MarketsandMarkets estimating a compound annual growth rate (CAGR) exceeding 47% between 2026 and 2032, reaching $15.24 billion. This expansion is driven by advancements in edge AI computing and real-time decision-making capabilities in robotic systems. Locally, entrepreneur Abhinav Kukreja recently launched Neocambrian AI, aiming to establish a high-fidelity, pre-training scale database of “human action from India,” mirroring Human Archive’s ambition. This indicates a growing trend of startups leveraging India’s vast labor market and diverse industrial landscape to build the foundational data for the next generation of physical AI.