An international
data corridor
for frontier AI.

Consented data from institutions in the USA, India and other countries, for pre‑training, post‑training and reinforcement learning.

Who we work with

We partner with researchers across industry, academia and frontier labs.

Dataset details are shared under NDA.

Request the data brief
01 · Data

Health first. Now across industries.

We started with health, where it matters most. Today we also offer enterprise workflow, design & CAD, customer service, gaming and security data, with early finance and legal cohorts.

Health

Available

12M+ health records across pathology, radiology and clinical care, including longitudinal, multimodal records from India’s leading tertiary hospital networks, US longitudinal cohorts and rare-disease genetic data. Live records from clinical systems, not claims or registries, following each patient across visits, admissions and years.

EMR / EHR · OMOP CDM Imaging · DICOM Pathology Biobank Omics
Explore health data →

Enterprise workflow

Available

Real enterprise workflows, showing how work actually gets done step by step. Built for training and evaluating AI agents.

Design & CAD

Available

Manufacturing and architecture design CAD data.

ManufacturingArchitecture

Customer service

Available

Studio-quality customer service data.

Gaming

Available

Consented gaming data.

Security

Available

Consented security data.

Finance

Early cohorts

An early finance cohort, growing as partners join.

Legal

Early cohorts

An early legal cohort, growing as partners join.

Need something else?

Tell us the data you need. We source new datasets through our partner network, consented and de-identified at the source.

Tell us what you need →

Expanding intoBusiness records · New cohorts as partners join

02 · Model training

Scoped project by project.

A single tranche or the full dataset, by domain, cohort or data type.

Pre-training

Large-scale, real-world records across domains.

Post-training

Targeted cohorts for specific domains, conditions or tasks.

Reinforcement learning

Grounded tasks built from longitudinal journeys and real enterprise workflows.

03 · Trust

Consent and privacy come first.

Consented
All of our data is collected with consent, through our partners’ own systems.
De-identified at source
Identifiable information is removed before data leaves the partner institution.
Under formal agreements
Every dataset is shared under formal agreements and NDA.
04 · Local to global

Local experience is what makes a model work everywhere.

Much of what today’s models learn comes from the open web, which reflects only part of the world. Institutional records carry local context: how experts actually work with the people, constraints and conditions in front of them.

Training on that context helps models feel local to the people using them, and approaches proven in one setting can carry to others. Local data has global impact.

“Globalization is not a product feature; it is a geopolitical negotiation.”
From The Trust Wall, Sakina’s essay on a16z.
05 · Founder & team
Twitch · Exec team YouTube Google 2× founder a16z New Media & Growth Engineering Fellow
Sakina Arsiwala

Sakina Arsiwala

linkedin.com/in/sakina

I’ve spent more than 20 years building and protecting products at internet scale. Most recently I was on the executive team at Twitch (Amazon), leading Community Health and Trust & Safety. Before that I was at Google and YouTube, and I’ve founded two companies.

Trust & Safety taught me that consent and privacy have to be built in from the start. Frontier models increasingly need data that sits inside institutions. I’m building a trusted route for that data to reach AI labs.

Previously at
Twitch Amazon YouTube Google

Team. Accomplished engineers and researchers from top technical schools like the Indian Institute of Technology (IIT), who have run pharmaceutical trials and evaluations with tertiary care hospitals.

Next step

Request the data brief.

We partner with researchers across industry, academia and frontier labs. Briefs for each dataset are available under NDA.