Clinical data, without the patient.
Medical records, notes and imaging prepared through licensing, de-identification and expert validation, linked at the patient level so multimodal records stay coherent.
Available as- 01 · Dataset
- 10M+ de-identified, patient-linked records
- 02 · Environment
- Clinical case tasks, clinician-graded
Licensed, stocked, ready to ship.
Plus custom collection for what doesn't exist yet.
De-identified medical records across specialties
CT, MRI, X-ray, ultrasound and pathology
Longitudinal, multimodal, across 15 countries
Multi-visit records with timestamped photographs
Maintained across projects
Including South and Southeast Asia, LATAM and the Middle East
Longitudinal, multimodal, linked.
Records travel with their context: imaging, notes, labs and samples stay linked at the patient level after de-identification.
- Medical imaging
- CT, MRI, X-ray, ultrasound and pathology slides for detection, segmentation, classification and tracking.
- Clinical records & notes
- Unstructured clinical text turned into usable data for clinical NLP, risk prediction and decision support.
- Oncology
- Whole-exome sequencing, digital pathology WSIs, FFPE tissue, DNA/RNA and structured records across 13 categories.
- Dermatology
- Clinic-sourced photographs tracking progression over multi-year follow-up, with redacted labs and prescriptions.
- Underrepresented populations
- A deliberate non-Caucasian focus, addressing a consequential gap in oncology and dermatology AI.
- Expert validation
- Healthcare professionals continuously validate annotations and flag inconsistencies.
De-identified record. Values are made-up examples.
From consent to clinician sign-off.
One note from patient_8841, through the five stages every healthcare record goes through before it ships.
Source & consent
Records come only from partners under agreement, with consent documented per patient for training use.
- Who
- Partner hospitals
- Gate
- No consent record, no ingest
De-identify
Names, record numbers, dates, places and contacts are found and replaced. Dates shift by a per-patient offset so intervals stay true.
- Who
- PHI model + reviewer
- Gate
- Every replacement human-checked
Link
Notes, imaging, labs and medications join on a hashed patient key, so the record stays one patient after the name is gone.
- Who
- Data engineering
- Gate
- Orphan items flagged, not dropped
Clinician review
Clinicians confirm diagnoses and findings and label what the task needs. Disagreements go to a second clinician.
- Who
- Clinicians
- Gate
- Low agreement is redone
Package
Shipped in the format your pipeline reads, with a datasheet describing sources, consent, de-identification and review.
- Who
- Delivery
- Gate
- Datasheet with every delivery
Built for frontier model tasks.
Scope, review criteria and delivery format are agreed per engagement, from the first sample to the final eval.
- Diagnostic imaging models
- Detection, segmentation and abnormality tracking.
- Clinical NLP
- Summaries, coding and extraction from clinical notes.
- Medical assistants
- Expert-reviewed dialogue and answers for clinical chat.
- Risk prediction
- Longitudinal records for outcome and progression models.
- Clinical evaluation sets
- Expert-graded benchmarks for medical reasoning.
What buyers ask us first.
Something else? Ask a specialist →
01How do you protect patient privacy?
We de-identify medical data and use access controls, documented handling procedures and project-specific contractual safeguards. HIPAA and GDPR requirements are addressed according to the data, processing roles and agreements for each engagement.
02What imaging data do you annotate?
CT, MRI, X-ray, ultrasound and pathology images, with labels for detection, segmentation, classification and abnormality tracking.
03Can you handle de-identified clinical records?
Yes. We prepare de-identified clinical records for annotation and AI workflows, with formats, fields and review steps scoped to the project. Samples are available on request.
04Do you have EHR data for LLM training?
Yes. De-identified patient records and clinical notes stay linked at the patient level with imaging and labs, and can be delivered as FHIR, HL7, JSONL or CSV for clinical NLP, summarisation and medical assistant training.
05Can you build clinical evaluation sets?
Yes. Clinicians write case-based questions with expert rubrics and a held-out answer key, and grade model answers the same way every run. Clinical case tasks can also be packaged as environments with clinician graders.
Same standard in every domain.
CAD & 3D
Drawings, assemblies and meshes with validated geometry.
- 01 Dataset
- 5M+ typed, native-format CAD files
- 02 Environment
- Part classification and rebuild checks
Audio & speech
Synchronized audio with transcripts and event labels.
- 01 Dataset
- 5.7M+ transcribed, diarized assets
- 02 Environment
- Transcription evals per language and accent
Egocentric video
First-person footage with action and object labels for physical AI.
- 01 Dataset
- 200K+ hrs synced video, IMU, action labels
- 02 Environment
- Action and manipulation benchmarks
Company operations
Workflows, approval trails and legal documents for enterprise agents.
- 01 Dataset
- 52+ companies' ordered approval trails
- 02 Environment
- Agent tasks over approvals, operator-graded
Conversational
Multi-turn conversations, ranked and reviewed for assistant training.
- 01 Dataset
- Multi-turn dialogue, intent-labelled
- 02 Environment
- Multi-turn evals, rubric-graded
Private code
Repositories, commits, pull requests and review history.
- 01 Dataset
- 3B+ lines with commit, PR and review history
- 02 Environment
- Agentic engineering tasks, engineer-graded
See real healthcare data first.
A representative sample under NDA, in the format your pipeline reads. Run it through your own evals first.
- Sample records, as delivered
- Datasheet and schema
- Review criteria
- DICOM · WSI · FHIR
{"case":"04-218","modality":"CT","series":"axial",
"note":"Patient [NAME], DOB [DATE], MRN [ID],
chest pain, referred [NAME]",
"identifiers":"removed","imaging":"linked",
"labels":{"finding":"nodule","segmented":true},
"reviewers":2,"status":"accepted"}