Research Engineer, ML Infrastructure

AI & Data · San Francisco office · FullTime · Senior level The role is about making large model training fast, stable and cost efficient. That means working on the training code from whole models down to individual GPU kernels, and keeping long runs on big GPU clusters healthy. A strong candidate has deep hands-on experience with distributed training, performance profiling, and reliability work such as checkpointing and recovery, plus solid Python and PyTorch or JAX skills. How it fits the AI era AI is the product here, so the work is about building the machinery that lets new model ideas be tested and scaled quickly. Tools that help write and tune code can speed up routine tasks, but deciding where a training run is losing time, and how to make it dependable, still takes careful human judgment. Working closely with researchers to turn new architectures into runs that scale stays a people-centered job. How it builds AI experience The role gives direct experience training frontier models on biochemical structure and interaction, which is a specialized area within AI for drug discovery. Someone can build toward it by practicing multi-GPU training of open protein or molecule models, profiling them, and fixing throughput and failure problems. Learning the basics of how these models are used by drug scientists helps connect the infrastructure work to real outcomes. Worth showing Distributed training at scale GPU profiling and optimization Fault tolerant training systems PyTorch or JAX depth This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role. About Chai Discovery Frontier AI foundation models and computer-aided design suite to predict and reprogram biochemical molecules, per the company. #J-18808-Ljbffr