Field notes · Clinical data for AI research
Data from real clinical care, for frontier models.
Mozu Health licenses datasets and builds custom evaluations for AI labs that train and evaluate models on healthcare tasks. The data follows a visit from the conversation through the clinic’s work to the records and outcomes that follow.
In brief
- What Mozu offers
- Ready-to-license clinical datasets, plus datasets, evaluations and benchmarks built to your specification.
- Who creates and grades it
- Medical doctors with experience in the dataset’s specialty. The company is led by an Emergency Medicine physician.
- How to request a sample
- Tell the team which healthcare data you need, using a company email. See how requests work.
Field note 01 · Illustrative encounter
One visit, three layers of data.
A clinic visit is more than a conversation. Mozu’s datasets capture care at three levels. The visit below is fictional and written for illustration. It contains no patient data.
Telehealth primary care · Adult patient, 40s
Fictional example09:02Visit
“Three days of cough, worse at night. A low-grade fever since Tuesday.” The physician asks about breathing, sleep and contacts.
Level 1The conversation between doctor and patient
09:18Clinic
A chest X-ray is ordered and a follow-up video visit is booked for Friday. The note is routed for co-signature.
Level 2The administrative work that runs the clinic
Day 4Record
The SOAP note is closed. At follow-up the fever has resolved and the cough is improving, and no further imaging is needed.
Level 3The patient records and outcomes that follow
Linking a decision in the visit to the work and the outcome after it gives a model more to learn from than the transcript alone.
Datasets and use cases
Built for training and evaluation.
Mozu datasets
Ready-to-license datasets drawn from real clinical care, with patient interactions, patient records, clinician decisions and outcomes.
Custom data development
Datasets, evaluations and benchmarks built to your specification, collected from clinical practice or created by experts.
Real-world data
Large de-identified patient record datasets that map SOAP notes to real-world care.
Use · Pretraining and evaluation
Reinforcement learning
Clinical and administrative tasks with feedback grounded in the decisions and outcomes that follow an encounter.
Use · Long-horizon clinical tasks
Benchmarks and evaluations
Tests designed around meaningful clinical work, typically by specialty or by type of clinic or hospital.
Use · Measuring a model by specialty or setting
Expert data
Created and graded exclusively by medical doctors with experience in the specialty of the dataset being built.
Use · Tasks that need clinical judgment
Specialties today
- Telehealth primary care
- Emergency Medicine
- Dermatology
- Psychiatry
- Japanese Medical Terminology
New specialties are added regularly.
Physician-led
Doctors make the data, and grade it.
Mozu’s founder and CEO, Mai Shimada, MD, is an Emergency Medicine physician trained in the United States and Japan. Physicians, engineers and repeat company builders make up the team in San Francisco and Tokyo.
Expert data is created and graded exclusively by medical doctors with experience in the specialty of the dataset being built.
Meet the founding teamMozu is the name of a bird known for its ability to communicate. We believe the words of patients from around the world are critical to improving AI models.
Request a sample
Tell us what you are studying.
- 1
Name the task
Training, evaluation or benchmarking, and the specialty or care setting you are studying.
- 2
Share a company email
The request asks for a company email and the healthcare data you are interested in.
- 3
Hear from the team
The team replies about ready-to-license datasets or custom data development.
Explore a dataset with Mozu.
Data to train and evaluate frontier models on healthcare tasks.