Software Engineer - ML Platform

About the Role

The Platform team at Replicate oversees the entire lifecycle of models, from packaging and deployment to serving, scaling, and monitoring. You’ll be developing the infrastructure that supports thousands of models and powers millions of predictions daily. This is a chance to build something truly innovative, where each decision you make has a tangible impact and allows your creativity to shine.

Responsibilities

• Designing and building our deployment and model-serving platform.
• Building technology to operate the latest advancements in the ML and AI space.
• Designing systems to maximize the utilization and reliability of our Kubernetes clusters and GPUs, including multi-regional traffic shifting and failover capabilities.
• Owning and optimizing fair and reliable task allocation and queuing across a diverse set of customers with heterogeneous workloads.
• Working with our Models team to speed up model inference through techniques like caching, weights management, machine configurations, and runtime optimizations in Python and PyTorch.
• Working with technologies such as: Python, Go, and Node.js; Kubernetes and Terraform; Redis, Google BigQuery, and PostgreSQL

Qualifications

• Experience building platforms at scale.
• Worked in complex systems with many moving parts; you have opinions on monoliths vs. services.
• Designed and implemented developer-friendly APIs to enable scalable and reliable integration.
• Hands-on experience setting up and operating Kubernetes.
• A passion for building tools that empower developers.
• Strong communication and collaboration skills, with the ability to understand customer needs and distill complex topics into clear, actionable insights. We believe that most of programming isn’t just about writing code; building a platform requires a collaborative approach.
• At least 3 years of full time software engineering experience.

Nice-to-have

• You have worked on machine learning platform teams in the past.
• You have experience working with or on teams that have put ML/AI into production, even though this role does not entail building ML models directly.
• You have some exposure to serving Generative AI features where GPUs are costly commodities and workloads can take significant time to finish.