Research Engineer, Voice AI [NYC or SF]
Concurrence
Get the Swapply iPhone app and it tailors your resume and prepares Concurrence’s application. If their site asks for a login or a human check, it comes back to you ready to finish, free. You pay only once an application is confirmed sent, and your first 3 are free.
or apply on Concurrence’s siteAbout the job
Locations in the US: New York, NY; San Francisco, CA
ABOUT CONCURRENCE
Concurrence (formerly Amigo) is building the platform healthcare organizations use to build their own clinical AI agents. Our mission is to make excellent care accessible to every patient, not rationed by the limits of clinician time.
Underneath our platform is the Health World Model, which unifies a health system’s data into a living, event-level record of every patient’s care. On top of it, agents handle work across the patient journey, from intake and scheduling to medication management and clinical decision support.
Our agents have supported more than 10 million patient encounters and are on track to grow tenfold this year. We run natively on Databricks, integrate with every major EHR, and are backed by Madrona, General Catalyst, and Optum Ventures. We’re about 35 people, working in person in New York and San Francisco.
ABOUT THIS ROLE
As a Research Engineer, Voice AI at Concurrence, you'll build and tune speech models for patient conversations and ambient clinical audio. You'll own models from data and training through evaluation and production, working with runtime engineers on streaming inference, latency and turn-taking. The work includes full-duplex speech-to-speech models that listen and speak at the same time, handle interruptions and preserve clinical details.
WHAT YOU'LL DO
- Train and fine-tune speech and audio models for patient calls and clinical encounters
- Own data preparation, training experiments and evaluation code
- Improve streaming inference, turn-taking and interruption handling, balancing model quality, latency and cost
- Define clinical failure cases with clinicians, product and evaluation engineers, and measure the errors that matter to patients
- Evaluate changes on held-out examples, investigate regressions and carry improvements into production
- Use evidence from real conversations to choose the next experiment and shape our modeling approach
WHAT WE'RE LOOKING FOR
- You've personally trained or fine-tuned speech or audio models on large audio datasets as a substantial part of your work
- You've shipped a model into a real-time product, with concrete results on streaming inference, latency, turn-taking or overlapping speech
- You've taken an ambiguous problem, defined a useful metric and led the work through to a model running in production
- You write the training and evaluation code yourself and can explain the choices behind an experiment
- You use coding agents in your work and independently verify their output
- You're low ego, direct, and hold yourself to a high bar
- You can work on site in New York City or San Francisco
NICE TO HAVE
- Experience building or tuning a full-duplex speech-to-speech model, our strongest preference
- Neural audio codecs or audio language models
- Telephone audio, overlapping speech or multi-speaker audio
- Model adaptation for a particular customer, domain or regulated environment
BENEFITS (AVAILABLE TO FULL-TIME EMPLOYEES)
HEALTH & WELLNESS
- Comprehensive health, dental, and vision insurance
- Daily catered lunch and dinner
- Mental health support and wellness coaching
- Flexible wellness stipend for fitness, therapy, or personal growth
GROWTH & DEVELOPMENT
- Annual learning budget for courses, books, or conferences
- Conference attendance budget for professional development
- Annual team offsite
- Academic collaboration opportunities
- Unlimited PTO
OUR CORE VALUES
- Patients Win, We Win
If patients aren't getting better care, we haven't earned the right to scale. Every internal decision gets pressure-tested: does this make patients' lives better? If we can't draw the line, we question why we're doing it.
- High Standards, High Care
We hold a high bar for the team because patients are counting on us to get this right. But high standards only work with genuine investment in each other. You can take risks, admit mistakes, and challenge ideas. Not despite our standards, but because of them.
- Thoughtful Urgency
We move fast by default, but speed without judgment is recklessness. The discipline is knowing which decisions are reversible vs. not. In healthcare AI, the companies that win will be fast everywhere they can be and careful everywhere they must be. We build the muscle to do both.
- Intensely Measured
We instrument patient outcomes, provider ROI, system performance, and clinical accuracy. But data without action is surveillance. Every metric should have an owner, a threshold, and a response plan. If we're measuring something but never acting on it, we stop measuring it.
WHO BUILDS WITH US
- Low ego: Politics and territory don't interest you. The best ideas win, regardless of who has them.
- Direct: You say the hard thing, challenge ideas openly, and commit fully once decided.
- High agency: You thrive on trust rather than instruction. When you see something is broken, you fix it. You don’t file tickets and wait for someone else.
- Bar of excellence: You hold yourself to a bar most people wouldn't, and you want teammates who do the same.
- Skeptical: You push back on rules that don’t make sense and question assumptions that haven’t earned their place.
Swapply is an independent app, not Concurrence. This job and its description come from Concurrence’s own careers page, where it was posted on November 16, 2026.