Your judgment is the part of the work a model can't scrape.
We employ clinicians, coders, adjusters, attorneys, engineers, and field technicians part time to teach and test AI systems built for their own professions. The work is remote, paid hourly at professional rates, and done on real records inside secure environments.
Why we need people who have done the job
A model can learn to sound like a nurse, an underwriter, or a litigator from public text. It can't learn when a prior-authorization summary leaves out the detail that changes the decision, or when a coverage letter cites the right exclusion for the wrong reason. Only people who have made those calls can teach that, and only they can tell whether a model has learned it.
Our clients are hospital systems, insurers, law firms, utilities, logistics operators, and the companies that build AI tools for them. They need professionals to write the examples their models learn from and to judge the answers those models give. That's the expert network.
You join as a part-time, hourly employee of Benner House, not as an anonymous gig worker. You're paid every two weeks by direct deposit, training and calibration time is paid at your full rate, and you see your own quality scores.
What the work is
Every project uses real-world material from your field, usually de-identified, presented in a secure workspace with a written guideline and rubric.
Rubric scoring
Read a model's answer to a real task, such as a discharge summary, a claim triage note, or a clause comparison, and score it against criteria you helped refine.
Preference judgments
Compare two answers to the same question, choose the better one, and say briefly why. These pairs teach models which of two plausible responses a professional would accept.
Writing demonstrations
Write the answer you would want from a capable colleague, step by step. Your demonstrations become the examples a model is fine-tuned on.
Reviewing model and agent output
Check a model's draft or an agent's sequence of actions, such as working a claim in a sandbox, and mark exactly where it went wrong and what should have happened instead.
How it pays
Rates depend on your credential, years of practice, and the project. Training, calibration, and qualification hours are paid at the same rate as production work.
| Domain | Typical roles | Hourly range |
|---|---|---|
| Medicine | Physicians, nurse practitioners, physician assistants | $95–$140 |
| Nursing and pharmacy | Registered nurses, pharmacists, clinical documentation specialists | $60–$95 |
| Medical coding | Coders holding CPC, CCS, or RHIT credentials; utilization reviewers | $45–$70 |
| Insurance | Claims adjusters, underwriters, subrogation specialists | $50–$85 |
| Law | Attorneys admitted in at least one US state | $100–$140 |
| Legal support | Paralegals, litigation support and e-discovery specialists | $45–$65 |
| Engineering | Licensed professional engineers: civil, electrical, mechanical, chemical | $85–$130 |
| Utilities and field service | Line workers, substation and meter technicians, field service technicians | $45–$75 |
| Logistics | Dispatchers, freight planners, terminal and yard coordinators | $45–$65 |
| Agriculture | Agronomists and crop advisers | $55–$90 |
| Accounting | CPAs, auditors, controllers | $70–$110 |
Requirements
These protect our clients' data and your professional standing. We check every one before you see a record.
- You live in the United States, are authorized to work here, and do all project work from within the US
- You hold an active license or credential in good standing where your field has one, and we verify it with the issuing board or body
- You have at least three years of hands-on practice in your field
- You pass a background check, which we arrange and pay for
- You sign a confidentiality agreement, plus a project-specific NDA where a client requires one, and complete HIPAA privacy training before any health project
- You can commit at least 8 hours a week for the length of a project, typically four to twelve weeks
- You work from a private space through the secure virtual desktop we provide, with no copying, photographing, or sharing of records
From application to your first project
Apply
Send your résumé or CV, your license type, state, and number if you hold one, your specialties, and the hours you could give each week.
A 30-minute call
An annotation lead in your domain talks with you about your practice: settings, specialties, jurisdictions, and the kinds of cases you've handled. We reply to every application within five business days.
A paid qualification task
One to two hours of items in the same format as real projects, scored against answers our senior reviewers agreed on. It's paid at your rate whether or not you continue.
Checks and onboarding
Background check, license verification, signed agreements, payroll setup, and access to the secure workspace. Most people finish onboarding within two weeks.
Matching and calibration
We match you by credential, specialty, setting, and jurisdiction: an emergency nurse isn't assigned oncology coding, and a Louisiana attorney isn't asked to judge a Mississippi filing. Every project starts with a paid calibration session before production work.
How quality is measured
The standards we promise clients apply to every expert, and you see your own numbers every week.
≥ 95%
accuracy on hidden gold questions
Items with agreed answers are mixed into every queue. Dropping below 95% pauses your queue for a paid recalibration session, not a dismissal.
≥ 0.80
Krippendorff's alpha
Agreement between experts on categorical labels, measured per batch. When agreement is low, we fix the rubric before we question the reviewers.
10%
independent second review
At least one item in ten gets a second reviewer. Disagreements go to adjudication, and the reasoning becomes a new guideline example.
Weekly
feedback
You see your gold accuracy, your agreement with peers, and reviewer notes on your work, so you can calibrate as you go.
How to apply
Email experts@bennerhouse.com with your résumé or CV, your license type, state, and number if you hold one, your specialties, and your weekly availability. Or use the contact form and choose the topic "Joining the expert network." Please don't send copies of patient, client, or employer records with your application. We reply within five business days.
Questions from experts
Do I need any experience with AI?
No. You need deep experience in your field and the patience to explain your reasoning in writing. We train you on the tools, the guideline format, and how scoring works, and that training time is paid.
Can I do this alongside my full-time job?
Yes, and most of our experts do. Work can be done in the evenings and on weekends within project deadlines. Check your employer's outside-employment policy first. We also screen out any project involving your current employer, its clients, or its direct competitors.
Will I see patient or client information?
Most material is de-identified before an expert sees it. When a project requires identifiable health information, the work happens inside the client's own environment under a business associate agreement, after HIPAA training, with access limited to the records the task needs.
How many hours are available?
It depends on client demand in your domain. Projects typically run four to twelve weeks at 8 to 20 hours a week. We'll tell you plainly when your domain is quiet rather than keep you waiting on a list.
Will my name be attached to the work?
No. Clients see the credentials and experience of the review team, not names, unless a client's contract requires named reviewers and you agree to it in writing for that project.
Am I helping build a model that replaces me?
The systems our clients build are mostly for drafting, summarizing, triage, and checking work, with a professional making the final decision. Your judgment is what makes them safe to use. If a project's purpose troubles you, you can decline it without any effect on future assignments.
Keep reading
Bring us one messy data source and one question.
Tell us what the data is and what you want a model to do with it. A data engineer, not a salesperson, will reply within one business day with questions and a proposed first step.