Research Scientist

Phare Health and R1 RCM

5d•$140,000 - $300,000•Hybrid

About The Position

Phare Health is now part of R1 and its AI innovation engine, R37 Lab, bringing Phare’s frontier clinical reasoning technology together with one of the largest healthcare platforms in the U.S. At R37 and Phare, we are building the first AI-native Healthcare Revenue Operating System: a connected platform that reasons over full medical records, payer logic, and financial workflows to automate medical coding, billing, and follow-up. Backed by real customers, real data, and real distribution, we operate on a national scale. Our agentic AI systems already power production workflows across 95 of the top 100 U.S. health systems, processing hundreds of millions of patient encounters each year, including: 180M+ Claims 550M+ Patient encounters 1.2B+ Workflow actions and outcomes each year This is startup-level ownership with enterprise-level impact. If you want to build AI that ships, scales, and measurably improves how healthcare works, this is the place to do it. We skew toward the RS/RE blend – applied scientists who feel comfortable getting close to production code. You’ll conduct applied research on real healthcare data, focused on explainability, reinforcement learning, and long-context information retrieval. You’ll move quickly from concept to deployment: designing experiments, training models at scale, and collaborating with ML Ops and Product teams to turn ideas into measurable user impact. You’ll stay close to the literature and close to production, coding your own experiments end-to-end. This is an in-person role in NYC, requiring at least 3 days in the SoHo office.

Requirements

You have a PhD in computer science / informatics and at least 3 years of industry experience (will consider postdocs).
Experience designing novel architectures and pipelines in PyTorch/TensorFlow/JAX - strong preference for applied research which has made its way into production
Research expertise in one or more of: interpretability, reinforcement learning, retrieval-augmented generation, or long-context information retrieval
Comfortable running large-scale training and evaluation on distributed infrastructure (e.g., Ray, FSDP, Lightning)
Track record publications at top ML venues (e.g., NeurIPS, ICML, EMNLP, MLHC, CHIL)