Software Engineer - ML Infrastructure
Epsilon Labs, Inc.2 months ago
Responsibilities
- Build and optimize distributed training infrastructure for foundation models on large-scale volumetric medical-imaging datasets.
- Build reinforcement-learning training infrastructure for rollout generation, reward-model serving, and experience collection.
- Develop high-throughput data loading, preprocessing, and multimodal data pipelines for production and offline datasets.
- Design centralized storage solutions with standardized formats such as protobufs for efficient retrieval and training.
- Partner with researchers to deliver production-ready systems from experimentation through deployment and monitoring.
- Contribute to model serving and deployment pipelines, including rollouts, canary deployments, and monitoring.
Requirements
- 5+ years building ML infrastructure, data pipelines, or production ML systems.
- Strong Python skills and expertise with PyTorch or JAX.
- Experience with distributed training at scale, including FSDP, DeepSpeed, or Megatron-style parallelism.
- Experience with data pipeline technologies such as Spark, Airflow, BigQuery, Snowflake, Databricks, or Chalk, along with schema design.
- Experience with distributed systems, AWS or GCP cloud infrastructure, and Docker or Kubernetes containerization.
- Track record of building scalable data systems and shipping production ML infrastructure.
- Preferred experience with reinforcement-learning training infrastructure, high-performance inference and serving, internal research experimentation platforms, A/B testing workflows, vision-language or multimodal architectures, DICOM and healthcare data standards, MLOps, and privacy-preserving systems with HIPAA compliance.
Tech Stack
Apache AirflowApache SparkAWSDatabricksDockerGoogle BigQueryGoogle Cloud PlatformKubernetesPythonPyTorchSnowflake
Categories
About Epsilon Labs, Inc.
Epsilon Health exists to solve the looming global radiology crisis before it reshapes patient care. Radiology underpins nearly every medical specialty, yet it is one of the most strained parts of the healthcare system. We’re rethinking how imaging and interpretation work to make radiology faster, more reliable, and future-proof.