Tech Stack

Description

You will work at the intersection of cutting-edge machine learning and rigorous research workflows to design and scale intelligent training systems for atomistic simulation models. Your role involves building foundations for training these models at scale, diving into architecture, data, optimizers, and representation learning to unlock their full potential. You'll help invent playbooks for pretraining foundation simulation models and contribute to transforming drug discovery through pioneering models that simulate the physical world with unprecedented speed and fidelity.

Requirements

• Experience working closely with ML researchers to turn scientific goals into engineering execution
• Designed training workflows that enable fast scaling, tracking results, and troubleshooting failures
• Knowledge of training aspects like learning rates, batch norms, weight initializations, and optimizer schedules
• Fluency in PyTorch and comfort with distributed cloud setups such as multi-node, multi-GPU
• Pragmatic DevOps skills, including familiarity with k8s, SLURM, or similar infrastructure
• Energized by uncharted problems and motivated to define new best practices in training world models
• Sense of relentless urgency and natural collaboration with a focus on team success
• Desire to work in a well-funded, bold, talent-dense organization on transformational impact

Responsibilities

• Scale FSM training by developing next-generation training pipelines for deep simulation models
• Map strategy by defining and iterating on short-, medium-, and long-term training strategies
• Engineer metrics by building robust training diagnostics and interpretability tools
• Debug at depth by diagnosing training failures and designing resilient, reproducible workflows
• Tune architectures by shaping and adapting them for improved training dynamics and performance
• Explore representations by investigating representation learning in the molecular data domain
• Automate workflows using generative coding tools to accelerate and automate processes