7 days ago
Responsibilities
- Design and operate training-as-a-service infrastructure for multi-GPU training jobs, metrics tracking, and completion notifications.
- Build CI/CD for models, including automated evaluation gates, canary rollouts, and A/B testing hooks.
- Own experiment tracking and data versioning for terabyte-scale datasets that change weekly.
- Monitor training job health, including GPU utilization, loss curves, and out-of-memory detection.
- Drive training cost governance through spot instances, preemptible VMs, and budget alerts.
- Partner with ML Scientists to translate reproducibility, experiment comparison, and checkpoint recovery needs into platform abstractions.
- Evaluate and integrate external model providers into the training and evaluation platform.
- Own platform direction and mentor engineers as the team scales.
Requirements
- Experience building or operating an ML platform and training orchestration system using Ray, Kubeflow, Airflow, or a custom solution.
- Experience with Docker and Kubernetes, including Jobs, CronJobs, and Helm.
- Proficiency in Python or Go for pipeline orchestration and infrastructure tooling.
- Production-scale experience with experiment tracking and data versioning tools.
- Ability to own platform direction, make platform investment decisions, and protect platform integrity.
- Experience mentoring engineers and operating with broad ownership across the ML lifecycle.
- Comfort partnering directly with ML Scientists and working in an early-stage, fast-moving environment.
Tech Stack
Categories
About SpreeAI
SpreeAI builds AI-powered virtual try-on, sizing, and styling tools for fashion retailers and brands, delivered through web and mobile integrations and a partner portal. The company sells SaaS and SDKs that embed photorealistic try-on into ecommerce sites, in-store displays, and clienteling apps to reduce returns and improve conversion. Founded in 2023 and headquartered in Los Angeles, it integrates with platforms like Shopify to onboard brand partners quickly.