Clinical histories
Encounter notes, treatments, diagnoses and follow-up records, with chronology and source references.
Healthcare data for model development
We work with data partners to qualify clinical and multimodal cohorts for post-training and evaluation, with the histories, tasks and evidence the research needs.
Hundreds of clinical-history environments
We build each environment from partner records after checking chronology, linkage and missing context, then run it against frontier models. Clinicians shape the rubrics.
Clinical and multimodal data
Imaging & pathology
Radiology studies, associated reports and digital pathology, with image-to-record linkage checked for each study.
Laboratory & molecular data
Lab observations, sequencing and molecular measurements, linked to the patient record.
Longitudinal cohorts
Records across visits and modalities, scoped around a research question.
Each task starts at a fixed point in the patient history
Future events stay out of the input. Each expected response has a rubric, evidence requirements and clinician review.
For extraction tasks, the model identifies a documented event, keeps its date and returns the supporting source span.
Inside a clinical environment
Patient history up to the task start
- Clinic visit
- Lab result, abnormal
- Medication started
- Imaging
- Clinic visit
- Task starts here
- Later events, hidden from the model
Care plan from the model
- MedicationChange, dose and reason
- Follow-upTest and timing
- ReferralSpecialty and urgency
Rubric from clinicians
- Acts on the abnormal resultCritical
- Avoids a known interactionCritical
- Sets follow-up timingImportant
- Cites evidence from the historyImportant
PassEvery critical item met, 3 of 4 overall
What each dataset includes
Cohort qualification
Population, time period, modalities, inclusion criteria and completeness, with gaps recorded alongside the coverage.
Rights and provenance
Source authority, permitted purpose and onward-use conditions, with source and delivery records kept connected.
Task and review design
The input boundary, reference evidence and evaluation rule for each task. Clinicians review the rubrics.
Delivery
Schema, formats and access path agreed with the receiving team. Raw source data and prepared training tasks ship separately.
Work with us
Tell us what the work is, what data it touches and what constraints apply. We’ll work out what to build and how to measure it.