Teams at Epsilon Health

Lead post-pretraining work on multimodal medical vision-language models: fine-tuning, reward modeling, RL (including RLHF), chain-of-thought and tool-use training, inference-time strategies, and publishable research to enable clinically grounded radiology report generation.

Build data filtering and curation pipelines for VLM and classifier training sets, develop model-based data-quality signals, design clinical evaluation methodology for radiology report generation, maintain clinical benchmarks, validate model-based evaluators against radiologist judgment, and implement continuous evaluation and regression testing across the research stack.

Build and maintain ML infrastructure and data systems for medical imaging: distributed training, RL training stack, high-throughput data pipelines, centralized storage, and production serving to accelerate research-to-production iteration.