Reinforcement Learning Scientist for Medical Imaging

Epsilon Health; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Epsilon Health is seeking a Research Scientist to advance post-training and reinforcement learning for multimodal medical imaging models to support radiology reporting and decision support. You will own the pipeline from supervised fine-tuning to reward modeling and inference-time strategy, leveraging clinical data and radiologist feedback.

You will work with large-scale multi-GPU systems, shaping model design, evaluation, and deployment while collaborating with clinical and engineering

Qualifications

• 6+ years of academia/industry experience in reinforcement learning, post-training, or multimodal ML.
• Strong foundation in post-training and reinforcement learning techniques including GRPO and related methods.
• Experience with reinforcement learning infrastructure at scale and multi-GPU training.
• Proficiency in PyTorch or JAX and ability to write production-quality code.
• Experience with model interpretability and safety-critical AI in healthcare contexts.

Responsibilities

• Design reinforcement learning with verifiable rewards for radiology report generation.
• Extend RL to noisy objectives using learned reward models and radiologist feedback (RLHF).
• Run GRPO-family algorithms with multi-reward objectives and diagnose reward hacking.
• Train multimodal reward models conditioned on images, with outcome and process supervision.
• Develop chain-of-thought reasoning over image regions with grounding verification loops.
• Train multimodal tool use (windowing, cropping, detector calls) with credit assignment.
• Develop inference-time methods including best-of-N sampling and verifier-guided decoding.
• Tune output style to reporting conventions, separating style vs content rewards.
• Stay current with RL, reward modeling, and multimodal post-training research.
• Drive publications and technical blogs to establish best practices in post-training medical VLMs.

Skills

Reinforcement learning

Post-training ML

Multimodal ML

PyTorch

JAX

Distributed training

RLHF

Inference-time decoding

Tools

verl

TRL

OpenRLHF

vLLM

Job description

Epsilon Health is seeking a Research Scientist to advance post-training and reinforcement learning for multimodal medical imaging models to support radiology reporting and decision support. You will own the pipeline from supervised fine-tuning to reward modeling and inference-time strategy, leveraging clinical data and radiologist feedback.

You will work with large-scale multi-GPU systems, shaping model design, evaluation, and deployment while collaborating with clinical and engineering