Attending Physician Expert Collaborator

About us

Alterra Labs is the safety and alignment company for AI in healthcare and the life sciences.

The most capable healthcare AI won’t be the model that gets the most benchmark questions right. It will be the one that can navigate the complexity of real clinical care without missing what matters. Healthcare decisions rarely happen in a clean sequence. Information is incomplete, records conflict, and the right action can depend on something that happened hours, days, or months earlier. A model can perform well on a test and still fail when it has to reason across a real workflow.

That’s why clinicians have to be part of how AI is evaluated, and why it has to happen now. AI is already entering the exam room, the inbox, and the chart. The people who decide what “safe” and “good enough” mean for these systems should be the people who have sat with patients, weighed uncertain evidence, and lived with the consequences of a decision. Being a physician has always meant two commitments at once: to heal the patient in front of you, and to advance what medicine knows for the patients who come after. Holding AI to a clinical standard is a continuation of both. “First, do no harm” applies to the tools we build, not just the care we give.

Alterra builds realistic simulation environments of real healthcare software: EHRs, imaging systems, payer portals, voice, and clinical trials. We seed them with real-world data and pair them with clinician-authored tasks and rubrics. Frontier AI labs, applied AI companies, health systems, and life sciences organizations use our work to see where their models break, why they break, and how to make them better before those failures reach patients.

Our name translates to “another earth.” We’re a team of over 60 clinicians, scientists, ML researchers, and engineers based in New York. We’re backed by Oak HC/FT, Flare Capital Partners, and Company Ventures.

About the role

Inside our environments, an AI model is given a realistic patient and a realistic clinical job. It has to search the chart, interpret labs and imaging, use clinical software, place orders, arrange follow-up, and respond when new or conflicting information appears. We don’t just check whether it reached the right answer. We look at how it got there: what it looked for, what it missed, which actions it took, and whether it recognized when it was uncertain. By our measure, a model that solves a task for the wrong reasons has failed it, because at the bedside it would have.

Every task is built by our clinical fellows (medical students, residents, and fellows) who design cases around the failures that matter most, and experts like you decide what good care looks like for each one. Some checks are simple: was the right patient selected, was the order placed? Others call for real clinical judgment: was the reasoning sound, was the risk weighed correctly, was this a meaningful error or an acceptable alternative?

What you find matters beyond the task in front of you. The failures our clinicians identify become the evidence that AI labs and health systems use to understand what a model can safely do, where it still needs human oversight, and how it needs to improve. As models get better, they fail at harder problems, and the work gets more interesting.

As an Expert Collaborator, your clinical judgment becomes the standard that AI is held to.

If you join, you will:

• Review and sign off on tasks. Clinical fellows (medical students, residents and fellows) build realistic clinical tasks. Working synchronously and asynchronously, you’ll review them for clinical accuracy and realism, confirm that the grading reflects real clinical judgment, and verify that model failures are clinically meaningful errors before anything ships. This is an iterative process that involves collaborators with the fellows and multiple rounds of refinement to be sure things are clinically accurate.

• Support fellows in real time. During your weekly live window, you’ll be available to fellows as they build tasks: answering clinical questions, pressure-testing ideas and helping them find the failure modes that matter.

• Bring your practice into the work. You’ll contribute the failure modes you see in real practice and help shape how we design and evaluate our work.

• Help build the company. Beyond core review, there are many opportunities to shape Alterra and the future of clinical AI safety and evaluation, from refining our review process to helping design our clinician events.

We’re looking for you if you have:

• Board certification and an active clinical practice, at least part time. If you’re currently in fellowship, you’re eligible if you’ve already passed your boards in another specialty and are practicing, or have practiced, in that specialty.

• Deep clinical judgment from real patient care: you know what matters in a messy chart, what can wait and what can’t, and when the guideline doesn’t fit the patient in front of you.

• A love of teaching. You enjoy helping medical students, residents and fellows sharpen their clinical reasoning, the same way your best attendings did for you.

• Curiosity about AI and a commitment to patient safety. No technical background required, just a belief that AI in medicine has to earn clinicians’ trust.

• A knack for clear, specific feedback, especially in writing, since much of the review happens asynchronously.

• Reliable availability for at least 10 hours a week during the two-week blocks you choose, including one 3-hour live window on a weekday.

• Comfort with ambiguity, at the bedside and beyond. You’re at ease with cases that have more than one defensible answer, and with a startup where processes are still taking shape.

• A builder’s mindset. You’re comfortable with a fast-moving startup and want to help shape how the work evolves, not just take part in it.

  This is a list of ideal qualifications. If you don’t meet every one of them, you should still consider applying. We encourage people from all backgrounds to apply.

How we work

Predictable two-week blocks. You commit to the two-week periods that fit your clinical schedule, and each block starts on a Monday. During the weeks you’re on, we ask for at least 10 hours of availability for task review.

Mostly asynchronous, with one live window. Most review happens on your own time: you’ll work through tasks, give written feedback and sign off when a task is ready. We also ask you to hold one 3-hour window each week, Monday through Friday, when fellows can reach you in real time. Together, our experts’ windows mean there’s someone available to help fellows every weekday.

Clinicians and engineers, side by side. You won’t be working in isolation. Clinical fellows build the tasks, experts like you set the clinical standard, and our ML researchers and engineers turn that judgment into rigorous evaluations. Every task goes through several sets of clinical eyes before it ships.

Built to fit a clinical life. Onboarding sessions are held on Sunday mornings so they don’t conflict with clinic, and each one falls right before a new two-week block. You’ll have your documents, accounts and tools a few days before your session, so you can start ready to go.

Always improving, together. We’re a growing startup, and our process will keep evolving as we learn. We’d rather shape those changes with our experts than hand them down, so bring the failure modes you see in practice, ideas for making review smoother, and anything else you think would make the work better.

Room to do more. Ten hours a week is the baseline, not the ceiling. If you want to go further, there are opportunities to join research discussions, help chart new directions for our evaluations, and bring the cases that keep you up at night straight into the work. Some of our most important ideas start with a clinician saying, “Here’s what I see in practice that no model would catch.” That could be you

Commitment: Experts join for an expected 3–6 month block. They do not have to work every 2 week block, but they need to be clear about when they can work. There are check-ins on fit and performance.

How to apply

After you complete this application you will get an email to with an assessment. You’ll need to download and complete an assessment that takes about 60 minutes. We’ll will review your application and get back to you about next steps