Job description

ABOUT US Epsilon Health is transforming medical imaging and diagnostics by bringing frontier AI into clinical practice. We're tackling some of healthcare's most critical challenges, building technology that directly improves patient outcomes. This year, we've served over 300,000 patients and now serve 3,000+ daily, working with some of the largest medical institutions in the U.S. By year-end, we're on track to interpret 1% of all daily U.S. X-rays. Our research team brings experience from DeepMind and Meta, powered by one of the industry's most comprehensive proprietary medical imaging datasets, continuously enriched by real-world clinical data. Together, we're building a future where frontier AI helps save millions of lives today, not tomorrow. Recently out of stealth, we're proudly backed by Jack Altman, Uncork Capital, Renegade Partners, and AlleyCorp, one of NYC's most active healthcare venture firms. ABOUT THE ROLE We're looking for an exceptional Research Engineer or Research Scientist to push the frontier of post-training and reinforcement learning for medical AI. You'll own the post-training of our multimodal foundation models, developing novel reinforcement learning methods, reward models, and reasoning capabilities that enable AI to generate clinically accurate and reliable radiology reports. You'll tackle fundamental research challenges in reinforcement learning, multimodal reasoning, reward modeling, and inference-time optimization across X-rays, CT, and MRI. Unlike traditional research labs, your work will draw on continuously growing real-world clinical data and feedback from practicing radiologists, creating a direct feedback loop between research and patient care. You'll work alongside a small team of researchers and engineers, with the freedom to pursue ambitious ideas and the resources to bring them to life. You'll collaborate closely with researchers responsible for foundation model pretraining. WHAT YOU'LL DO - Advance the frontier of reinforcement learning. Develop and scale novel post-training methods for multimodal foundation models, including reinforcement learning with verifiable and learned rewards. - Build better reward models. Design reward functions and train multimodal reward models using clinical accuracy, visual grounding, expert preferences, and real-world radiologist feedback. - Push the limits of multimodal reasoning. Develop methods that improve how models reason over complex medical images, localize evidence, and generate clinically grounded reports. - Explore agentic AI and tool use. Train models to interact with imaging tools, retrieve relevant information, and reason across multi-step workflows. - Scale inference-time intelligence. Investigate test-time compute, verifier-guided generation, and other techniques to improve model accuracy and reliability. - Own research end to end. Take ideas from hypothesis and experimentation through large-scale distributed training, evaluation, and deployment handoff. - Define how progress is measured. Develop rigorous evaluations for clinical accuracy, factuality, visual grounding, and model reliability. - Contribute to the research community. Publish original research, share technical insights, and help shape the direction of medical AI. WHAT WE'RE LOOKING FOR We care more about what you've built and discovered than where you've worked or how many years you've spent doing it. - Deep expertise in reinforcement learning, language model post-training, or multimodal machine learning. - Demonstrated ability to develop, train, or improve state-of-the-art machine learning models. - Strong understanding of modern post-training techniques, including reinforcement learning, reward modeling, and preference optimization. - Experience with large-scale distributed training using PyTorch or JAX. - Ability to independently formulate research questions, design experiments, and turn promising ideas into working models. - Strong engineering skills and a track record of implementing and extending ideas from research. - A rigorous, experimental mindset and the ambition to solve difficult, unsolved problems. NICE TO HAVE - Publications at leading ML conferences such as NeurIPS, ICML, ICLR, CVPR, or ACL. - Experience with reinforcement learning algorithms such as GRPO, DAPO, or related policy optimization methods. - Experience developing reward models, RLHF pipelines, or reinforcement learning with verifiable rewards. - Familiarity with large-scale RL infrastructure, including vLLM, SGLang, verl, TRL, or OpenRLHF. - Experience with multimodal reasoning, visual grounding, or agentic reinforcement learning. - Experience with inference-time scaling, verifier-guided decoding, or test-time optimization. - Background in medical imaging, radiology, or other scientific imaging applications. - Experience evaluating model factuality, hallucinations, or reliability in high-stakes settings. We welcome exceptional researchers from both academic and industry backgrounds. A traditional career path or prior healthcare experience is not required.

Requirements

Must-haves

We're looking for an exceptional Research Engineer or Research Scientist to push the frontier of post-training and reinforcement learning for medical AI.
You'll tackle fundamental research challenges in reinforcement learning, multimodal reasoning, reward modeling, and inference-time optimization across X-rays, CT, and MRI.
Unlike traditional research labs, your work will draw on continuously growing real-world clinical data and feedback from practicing radiologists, creating a direct feedback loop between research and patient care.
Advance the frontier of reinforcement learning. Develop and scale novel post-training methods for multimodal foundation models, including reinforcement learning with verifiable and learned rewards.
Build better reward models. Design reward functions and train multimodal reward models using clinical accuracy, visual grounding, expert preferences, and real-world radiologist feedback.
Push the limits of multimodal reasoning. Develop methods that improve how models reason over complex medical images, localize evidence, and generate clinically grounded reports.
Explore agentic AI and tool use. Train models to interact with imaging tools, retrieve relevant information, and reason across multi-step workflows.
Scale inference-time intelligence. Investigate test-time compute, verifier-guided generation, and other techniques to improve model accuracy and reliability.
Own research end to end. Take ideas from hypothesis and experimentation through large-scale distributed training, evaluation, and deployment handoff.
Define how progress is measured. Develop rigorous evaluations for clinical accuracy, factuality, visual grounding, and model reliability.
Contribute to the research community. Publish original research, share technical insights, and help shape the direction of medical AI.
We care more about what you've built and discovered than where you've worked or how many years you've spent doing it.
Deep expertise in reinforcement learning, language model post-training, or multimodal machine learning.
Demonstrated ability to develop, train, or improve state-of-the-art machine learning models.
Strong understanding of modern post-training techniques, including reinforcement learning, reward modeling, and preference optimization.
Experience with large-scale distributed training using PyTorch or JAX.

Nice-to-haves

Publications at leading ML conferences such as NeurIPS, ICML, ICLR, CVPR, or ACL.
Experience with reinforcement learning algorithms such as GRPO, DAPO, or related policy optimization methods.
Experience developing reward models, RLHF pipelines, or reinforcement learning with verifiable rewards.
Familiarity with large-scale RL infrastructure, including vLLM, SGLang, verl, TRL, or OpenRLHF.
Experience with multimodal reasoning, visual grounding, or agentic reinforcement learning.
Experience with inference-time scaling, verifier-guided decoding, or test-time optimization.
Background in medical imaging, radiology, or other scientific imaging applications.
Experience evaluating model factuality, hallucinations, or reliability in high-stakes settings.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aimachine learningml

Platform: ashby
Effort: quick