Human Archive

Problem / Solution Report

Problem Robotics models have traditionally relied on simulated environments or small, lab‑collected datasets. These sources lack the diversity, noise, and edge‑case scenarios found in real‑world settings, creating a "robotics data bottleneck" that hampers the transition of models from controlled labs to unstructured real‑world workplaces and homes.

Solution Human Archive addresses this gap by "archiving the physical world for embodied intelligence." The company deploys custom hardware across residential, manufacturing, and warehouse environments to capture egocentric video, depth, hand/body pose, force‑feedback, and teleoperation data. Collected streams undergo internal QA, anonymization, and multimodal annotation pipelines, producing high‑fidelity datasets that can be directly consumed by frontier robotics labs and general‑purpose robotics companies.

Value Proposition

• Real‑world fidelity: Datasets reflect true environmental variability, reducing the simulation‑to‑reality gap.
• Vertical integration: End‑to‑end pipeline—from field collection to labeled data—ensures consistency, compliance, and rapid iteration.
• Scalability: A distributed labor force operating in Asia, the U.S., and Europe enables large‑scale, diverse data acquisition.
• Accelerated development: Customers can train foundation models faster, lower on‑robot trial‑and‑error costs, and bring products to market sooner.

Impact By supplying multimodal, aligned data, Human Archive empowers robotics teams to build models that generalize across domains, improve safety, and achieve higher performance in real‑world tasks. This capability is increasingly critical as the robotics market expands and as embodied AI moves toward general‑purpose autonomy.

Market & Competitors Report

Market Context & Trends

• The global AI data collection & labeling market was USD 3.77 B in 2024, projected to exceed USD 17 B by 2030 (Grand View Research). Image/video workloads dominate, directly relevant to robotics data.
• Parallel growth in industrial (≈USD 34 B in 2024) and service robotics (≈USD 47 B in 2023) creates rising demand for high‑quality training data as companies shift from simulation to real‑world data.

Competitive Landscape

• Direct data‑collection peers:
  • Cortex AI (YC F25) – Provides large‑scale, real‑world workplace egocentric and robot datasets, overlapping heavily with Human Archive’s value proposition.
  • Qualitest – Offers human‑driven egocentric data capture via wearables, an adjacent service.
  • SunTec.ai – Supplies robotics‑focused training datasets and annotation services.
• Data‑labeling platforms (indirect competition):
  • Scale AI, Encord, Segments.ai, SuperbAI, BasicAI, Kili Technology, CVAT, Sama, Appen, Labelbox – Offer multimodal annotation tools and workforces but generally lack the upstream field‑collection capability that differentiates Human Archive.

Positioning & Advantages Human Archive’s key differentiation is its vertically integrated pipeline that owns both the upstream data capture (custom hardware deployed globally) and downstream QA/annotation. This reduces domain gaps and provides compliance controls absent in pure labeling platforms. Backing from Y Combinator and rapid operational rollout across Asia and the U.S. bolster credibility.

Risks & Considerations

• Strong competition from well‑funded labeling platforms (e.g., Scale AI) that could expand into field collection.
• Some robotics labs may choose to build internal data collection capabilities, limiting outsourced spend.
• Operational challenges around privacy, safety, and logistics at scale could act as barriers to entry for competitors but also pose execution risk for Human Archive.

Conclusion Human Archive operates in a growing niche at the crossroads of AI data services and robotics. While the market size is already multi‑billion dollars and accelerating, the company’s unique end‑to‑end approach and early YC backing position it competitively against both direct peers and broader data‑labeling platforms.

Founders' Background Report

Human Archive was founded by Rushil Agarwal, Samay Maini, Raj Patel, and Shloke Patel. All four are alumni of top U.S. universities, with several having dropped out of Stanford or UC Berkeley to focus on the venture. The Y Combinator profile notes that the team “dropped out of Stanford and Berkeley and moved to Asia” to build the world’s largest annotated multimodal dataset for embodied intelligence.

• Rushil Agarwal – Co‑founder. Educated in UC Berkeley’s M.E.T. program (Industrial Engineering & Operations Research + Business). His public posts emphasize solving the “robotics data bottleneck” by creating high‑fidelity multimodal datasets.
• Samay Maini – Co‑founder. Holds a B.S. in Electrical Engineering and Computer Science from UC Berkeley. Experience includes machine‑learning work at Lightning AI, a venture stint at Critical, and a software engineering role at Amazon in 2025. He focuses on transitioning from simulation‑based training to real‑world data capture.
• Raj Patel – Co‑founder. Described as a Berkeley dropout who previously worked as a farmer. His narrative centers on the need for real‑world data to enable robots to “see, move, and feel like humans.”
• Shloke Patel – Co‑founder. Listed by Y Combinator as “building in robotics.” He helped design and deploy custom hardware for data collection across Asia and the United States.

Collectively, the founders combine deep academic roots, hands‑on ML/AI experience, and operational expertise in field data collection, positioning them well to address the data scarcity in robotics.