# K01

> A research lab generating private, realistic synthetic patient trajectories for medical research.

K01 (K01 ehf.) builds generative models that learn the patterns in real clinical data without memorising the people in it. Training is differentially private, so a model carries the signal, not the individuals, and can produce synthetic patient data realistic enough to work with while no real patient information is exposed. Based in Reykjavík, Iceland (EU/EEA).

## Mission

Making health computable.

## What K01 does

K01 generates synthetic patient trajectories: multi-modal (demographics, conditions, medications, observations) and longitudinal (over time). You describe the cohort you need and the platform generates matching synthetic records on demand, so researchers, clinicians and developers can work with data they otherwise could not access.

How it works:
- Models are trained on real clinical data.
- Because training is differentially private, the trained model carries the patterns, not the people, and cannot leak any individual.
- Before release, the model is validated on three pillars: statistical similarity to real data, consistency with known clinical rules, and downstream utility.
- Users then generate cohorts conditionally through the API.

## Capabilities

- Differential privacy guarantees (no individual can be re-identified)
- Multi-modal, longitudinal patient trajectories, not static snapshots
- Conditional generation by cohort
- FHIR R4 and R5 output, plus CSV
- Reproducible by seed
- REST API, with an MCP server for natural-language queries
- Cloud or on-premise deployment (on-premise is a custom engagement)

## What K01 is used for

- Data for cohorts you do not have: a population you lack, a market you are entering, the rare cohorts and edge cases real records lack
- Designing and simulating trials before the first enrolment
- Internal research without waiting on data-access approvals
- Sharing reproducible cohorts across institutions

## Compliance

- GDPR: European data residency in Iceland (EU/EEA); no real patient data on production systems
- EU AI Act: synthetic data carries no personal data, which eases the data-governance burden on medical AI development
- HIPAA: synthetic data is not PHI

## Research

K01 publishes its methods and submits its claims to peer review.

- "The Viability Boundary of Differential Privacy", ICLR 2026 DATA-FM Workshop (https://openreview.net/forum?id=Pv3PSfaphM)
- "Tensorised Modular Architectures for Multi-Omics Generation", ICLR 2026 Gen² Workshop (https://openreview.net/forum?id=dgLH2Kuw50)
- "Multimodal Alignment for Synthetic Clinical Time Series", EurIPS 2025 Multimodal Representation Learning for Healthcare Workshop
- "Transparent Reporting for Healthcare GenAI", NeurIPS 2024 GenAI4Health Workshop (https://openreview.net/forum?id=cHnpUBShJP)
- "Classifying GenAI under the EU Medical Device Regulation", NeurIPS 2024 GenAI4Health Workshop (https://openreview.net/forum?id=glTUV6Uvqy)

## Team

- Dr Arinbjörn Kolbeinsson, CEO & Co-founder. Machine learning and clinical inference; previously Samsung AI and Evidation Health. PhD from Imperial College London.
- Dr Benedikt Kolbeinsson, CTO & Co-founder. Computer vision and large-scale synthetic data for ML. PhD from Imperial College London.

## Contact

- General: hello@k01.is
- Security: security@k01.is
- Careers: careers@k01.is

## Links

- Website: https://k01.is
- Security & Compliance: https://k01.is/security
- Startup Program: https://k01.is/startup
- API Documentation & Status: https://status.k01.is