Eden is the largest cloud native and AI native medical imaging OS. Our platform is used by 2,500+ imaging centers across 18 countries. We see 20M+ patients per year, and over 4 billion medical images spanning diverse populations that are chronically underrepresented in typical training sets. Thousands of Radiologists use our software every shift. This isn't a research prototype.
We're training multimodal foundation models for radiology. Eden is the diagnostic system of record at the centers we serve. That lets us train on all the clinical information that goes into a radiologist's read, and ensures every model output is reviewed and signed off by a radiologist during normal clinical work. That's a feedback loop most research teams don't have access to.\
\
**What you'll work on:** VLM architecture, training, and evaluation for radiology workflows. Clinical validation and dataset curation at a scale most researchers never get access to.
**_Technical_**
* You've trained non-trivial VLMs or LLMs end-to-end. Pretraining, large-scale SFT, continued pretraining, or post-training (RLHF / DPO / GRPO). Research or production both count. Inference on public checkpoints or lightweight LoRA tuning doesn't.
* You know what it takes to run large experimental campaigns on frontier-scale models: distributed training with FSDP or DeepSpeed, FlashAttention, mixed precision, activation checkpointing, data loading that keeps GPUs fed, eval that doesn't waste cycles. You have opinions about where the bottlenecks usually are.
* PhD in CS, EE, or a related quantitative field preferred. What you've shipped and published matters more.
* Experience with medical imaging or DICOM/PACS is a bonus, not a requirement.
**_General_**
* You want to know the clinical failure modes of your model, not only its test set numbers.
* You're comfortable sitting with radiologists and turning clinical problems into ML problems.
Visa sponsorship is offered for this role.
