Machine Learning Engineer (Infrastructure / Inference)

Machine Learning Engineer (Infrastructure / Inference)San Francisco, California (Hybrid)$170,000 - $220,000 + Equity + Healthcare + 401(k) + PTOAre you an ML Engineer looking to work on challenging AI infrastructure and inference problems while joining a high-growth AI start-up building systems that enable frontier multimodal AI to operate reliably at production scale?This is an opportunity to join a fast-growing AI company developing next-generation multimodal systems that require infrastructure beyond traditional backend engineering. The team works on complex problems involving GPU orchestration, large-scale inference systems, performance optimization, and developer platforms that allow applied scientists to move fast without sacrificing reliability or cost efficiency.In this role, you will create the foundational systems required to scale advanced AI technology safely and efficiently. You will work closely with applied science and infrastructure teams to bring research from experimentation into scalable production systems.This role would suit an engineer who enjoys operating at the boundary between systems engineering and machine learning while building production-ready infrastructure.The Role:*Architect scalable backend systems for training and inference pipelines. *Develop high-throughput APIs to facilitate efficient model serving. *Optimize GPU utilization and latency for complex multimodal workloads. *Construct distributed environments for large-scale generative models. *Enhance the observability and operational stability of AI architectures. The Person:*Proficiency in object-oriented programming (Python, C++, Java, Go, or related). *Proven command of core data structures and algorithm design. *Background in delivering production-grade backend or distributed software systems. *Working knowledge of cloud-native infrastructure and containerization strategies. *Experience with GPU-accelerated ML environments or specialized model serving frameworks like Triton, vLLM, or Ray Serve