Software Engineer - ML Infrastructure

The Opportunity

We are looking for an ML Infrastructure Engineer with 6+ years of experience to come in and establish best practices for our distributed training, reinforcement learning, and inference systems. You'll be a senior, autonomous engineer who can teach the team what world-class ML infrastructure looks like — someone who's operated at scale at a top-tier engineering organization and is ready to own the systems that power our foundation models. This is a ground-floor opportunity at a 5-person team backed by Bedrock Capital, Trust Ventures, and Jack Altman, building the first radiology-as-a-service platform powered by a multi-modal AI copilot to solve a crisis that leads to 200k deaths per year.

What You'll Be Doing

• Own and build the distributed training infrastructure for foundation models on large-scale medical imaging, including parallelism and checkpointing for volumetric data
• Design and operate the reinforcement learning training stack — high-throughput rollout generation, reward-model serving, and experience collection at scale
• Partner directly with researchers (ex-DeepMind, ex-Meta) to understand their workflows, prototype new ideas, and translate them into production-ready systems
• Build high-throughput data loading and preprocessing pipelines that keep GPUs saturated on large volumetric and multimodal datasets
• Contribute to production serving and deployment pipelines, including model rollout, canary deployments, and monitoring
• 6+ years of experience as an ML infrastructure engineer or distributed systems engineer

About the Company

The company is a trailblazing startup, setting new standards in healthcare through our innovative technology solutions. We specialize in integrating AI with medical imaging, aiming to improve diagnostic accuracy and patient care.

We are building the first radiology-as-a-service solution powered by a multi-modal radiology copilot. There is a dire radiology shortage (1 radiologist for every 4 radiologist jobs) that is leading to misdiagnosis with horrible consequences (200k people dying a year) - we’re going to solve that.

We’ve raised from Bedrock Capital, Trust Ventures, Jack Altman. We already have access to over 250,000 unique radiology images to train our initial mode on. This includes X-rays, MRI’s, CAT scans, echo’s, and ultrasounds. We also have our first customer secured (signed a $300k annual contract, not just an LOI).

Founded in 2023, team size around 5 employees, $1.6M in total funding raised.

About the Team

Founder - https://www.linkedin.com/in/rustinrassoli/ Founding CV engineer - https://www.linkedin.com/in/kedar-grama/ Founding ML Engineer - https://www.linkedin.com/in/ruian0/ Fractional VP of Eng - https://www.linkedin.com/in/bryanhall/ Technical advisor - https://www.linkedin.com/in/serkankutan/

The Product

The company is building the first radiology-as-a-service platform powered by a multi-modal radiology copilot. The platform integrates AI with medical imaging — spanning X-rays, MRIs, CAT scans, echocardiograms, and ultrasounds — to improve diagnostic accuracy and speed for healthcare providers facing a critical radiologist shortage. The product connects directly into hospital and clinic workflows through integrations with PACS, RIS, and EHR systems using medical communication protocols like DICOM and HL7. It enables radiology case distribution, AI-assisted image interpretation, radiologist peer review, and business analytics — effectively allowing healthcare organizations to scale their radiology capacity without needing to hire scarce radiologists. The company has access to over 250,000 unique radiology images to train its initial models and has already secured its first customer with a $300K annual contract, demonstrating early product-market fit in a space where 1 radiologist exists for every 4 open positions and an estimated 200,000 people die annually from misdiagnosis.

Tech Stack

Python, PyTorch, JAX, Kubernetes, FSDP, DeepSpeed, Megatron, vLLM, SGLang, TensorRT, Triton, CUDA, Docker

---

Application Questions

1. Are you able to work in San Francisco and come into the office 5 days per week?
2. Walk us through your experience building distributed training infrastructure (e.g. FSDP, DeepSpeed, or Megatron-style parallelism) and the scale you've operated at.
3. Describe your experience building ML infrastructure or platforms that research teams rely on for experimentation and training workflows.