Human Archive

About Human Archive

Human Archive is a data and deployment infrastructure company for Physical AI. Its work supports researchers and teams building systems that need to understand how people move, interact and perform physical tasks.

The company makes multimodal datasets available across RGB, RGB-D, audio, tactile, motion-capture and robot data. That mix creates an infrastructure problem as much as a research problem: large inputs need to be collected, organised, annotated, processed on GPU systems and made usable without losing track of access or provenance.

What needed building

Human Archive needed one operating path across cloud infrastructure, compute and data work. Treating each as a separate project would have left expensive GPUs waiting on data, annotation teams working around infrastructure gaps and cloud access growing by exception.

• GPU operations: Repeatable compute environments for workloads whose capacity and runtime needs differ from ordinary web services.
• AWS foundations: Managed infrastructure for networking, storage, compute, identity, monitoring and day-to-day change.
• Dataset movement: A dependable path for large, heterogeneous source data to move through processing and annotation.
• Annotation operations: Platform workflows for task assignment, annotator access, quality review and output tracking.
• Security: Clear boundaries between operators, annotators, workloads and the datasets they are allowed to reach.
• Cost visibility: Enough ownership and usage context to understand GPU, storage and transfer spend.

The operating architecture

We managed the AWS and security foundations around the full data path, rather than optimising GPU machines or the annotation interface in isolation.

What Gritt managed

GPU systems

We set up and operated the compute foundations needed for GPU-backed data and model workloads. The focus was repeatability: environments could be provisioned with known dependencies, connected to the right data and observed as real infrastructure rather than treated as special research machines.

AWS infrastructure

We managed the AWS foundation surrounding those workloads, including compute, storage, networking, identity and operational visibility. Infrastructure changes were handled as a connected system so a new processing workload did not require a parallel set of manual access and network decisions.

Cloud security

Security work focused on who and what could reach datasets, annotation workflows and compute environments. Human identities, workload identities and administrative paths needed different access, with logs and ownership available when something changed.

Data-annotation platform

We helped establish the infrastructure behind annotation operations: getting source material into the workflow, controlling annotator access, representing task state, supporting review and keeping completed outputs connected to the datasets they belonged to.

Technical priorities

The important design choice was to keep the data path and the infrastructure path connected. Annotation output had to remain traceable to its source; GPU workloads needed controlled, high-throughput access to the right material; and operators needed enough telemetry to distinguish a code issue from a capacity, storage or data-quality problem.

What changed

The engagement gave Human Archive a shared infrastructure base for data operations and compute instead of a collection of one-off environments.