Clinical data written by practicing physicians.
Medical models fail in the details a clinician notices at once. We commission vetted physicians and clinicians to write, label and grade the clinical data those models learn from.
Clinical medicine, with proof.
General-purpose models now pass licensing-style multiple-choice exams, yet still miss what matters at the bedside: a red-flag symptom buried in a long history, an interaction with a medication the patient already takes, a differential that stops one diagnosis short. Closing that gap takes data written by people who have made those calls. HUMXN sources clinical cases, reasoning traces, preference judgments and evaluation items from credential-checked physicians and clinicians, written to your specification and reviewed by a second expert before delivery.
Every clinical item arrives with its provenance attached. Documents and tables are scanned for personal information and masked where needed, items are fingerprinted at intake and checked for duplicates, and each label records whether it came from the author, was suggested by a model, or was confirmed by an expert reviewer. You receive a signed receipt listing every item, and an Ed25519-signed audit trail you can show to a compliance or clinical safety team.
For physicians and clinicians, this is paid, remote work in your own specialty. You write cases, grade model answers against a rubric and review colleagues' work, at a rate shown before you accept each job. The work is data creation and review for AI systems. It does not involve treating patients, advising members of the public, or forming a clinician-patient relationship. Your identity is not published, and approved work is paid automatically through Stripe.
Built to your specification.
Every engagement starts from a written spec and a pilot batch. These are the most common requests we source for clinical medicine.
Clinical cases and vignettes
Original cases written by physicians in the relevant specialty, with history, examination, investigations and a reasoned management plan.
Diagnostic reasoning traces
Step-by-step differential diagnosis and treatment reasoning, written the way a senior clinician would explain it to a resident.
Preference and RLHF judgments
Clinicians compare model responses for accuracy, safety, completeness and appropriate escalation, with a written rationale for each choice.
Evaluation and benchmark sets
Held-out question sets with rubrics, graded answers and specialty tags, newly written so they have not appeared on the open web.
Clinical note annotation
Labels for problems, medications, findings, negation and timing in de-identified notes, with PII scanning and masking on every document.
Safety and red-team items
Prompts that probe dangerous advice, missed emergencies and overconfident answers, with the expected safe behavior written by a clinician.
01Fluent is not correct
Models produce confident answers a clinician would reject. Only domain experts can grade the reasoning as well as the final answer.
02Contaminated exam data
Public question banks are widely scraped. Newly written, fingerprinted items give you evaluation results you can rely on.
03Patient privacy
Clinical text carries identifiers. Documents are scanned and masked, and nothing is sourced without consent bound to the item.
04Accountable labels
Label provenance and second-expert review show who made each judgment and who confirmed it, which matters when a model touches care.
Paid work in your field.
Remote and flexible, with the rate stated before you accept. Every job is reviewed by a second expert, and approved work is paid automatically.
- MD, DO, MBBS or an equivalent medical degree
- A current or recent license to practice in your jurisdiction
- Completed residency or specialty training, or equivalent clinical experience
- Several years of direct patient care in your specialty
- Nurse practitioners and physician assistants with relevant specialty experience
Write clinical cases
Draft realistic cases with a reasoned answer and the decision points a model should get right.
Grade model reasoning
Score model responses against a rubric for accuracy, safety and escalation, and explain each deduction.
Annotate de-identified notes
Label problems, medications, findings and negation in clinical text, following a written guideline.
Review a colleague's labels
Act as second reviewer on another clinician's items, confirming or correcting them before delivery.
Clinical medicine: asked often.
Can we get physician-written data in a specific specialty?
Yes. Requests are scoped by specialty, case mix and difficulty, and matched only to experts whose credentials are verified in that field. A cardiology evaluation set is written and reviewed by cardiologists, not generalists. We usually start with a pilot batch so your team can check the clinical bar, the rubric and the format before the request scales.
Is this work practicing medicine?
No. The work is creating and reviewing data for AI systems. You do not treat patients, advise members of the public or form a clinician-patient relationship, and nothing you write is given to an individual as medical advice. You apply your clinical knowledge to cases, questions and model outputs, following the written specification for each job.
Do I need a current medical license?
Most clinical work requires a medical degree and a current or recent license, which we verify along with your identity. Some tasks, such as annotating notes or checking factual content, may be open to clinicians with other qualifications. Each job states the credentials it requires, and you are only offered jobs that match what we have verified.
How do I get paid, and how much time is required?
Rates are set by field and task and are shown before you accept a job. You choose which jobs to take and work remotely on your own schedule, within the stated deadline. Once a second expert approves your work, it is credited to your balance and paid out automatically through Stripe after your identity and tax details are verified.
How do you handle patient data?
Most clinical data we deliver is newly written by clinicians, so no real patient is involved. Where documents or tables are sourced, they come from consenting sources, are scanned for personal information and masked where needed, and each item is bound to its consent record. Every item is fingerprinted and listed on a signed receipt.
Can we license the data exclusively?
Yes. Commissioned clinical datasets can be licensed exclusively, so the same cases and evaluations are not offered to anyone else. Scope, term and exclusivity are agreed in writing before sourcing begins. Deliveries come in JSONL, CSV, Parquet or Croissant, with a signed receipt listing every item by ID and hash.
Clinical medicine, done by experts.
Need expert-made data in this field, or have the expertise to make it? Start here.