Senior Machine Learning Engineer

hum.ai•San Francisco, CA

About The Position

Hum is building a world model to represent the dynamics of Earth processes, forecast how places change, and make that understanding useful. As Senior Machine Learning Engineer, you will own the software and infrastructure used to train and serve the company’s models, enabling research scientists to focus on model development. You will build and maintain the path from ingested data to training, systems for experimentation and large-scale training, and the inference path that delivers model outputs to the agent. You will collaborate with engineers and AI research scientists to make models trainable, measurable, and serviceable at scale, from initial experiments through full training runs.

Requirements

Bachelor's degree in computer science, engineering, or a related field, or equivalent experience.
At least 5 years of relevant industry experience, including hands-on machine learning engineering.
Proficiency building distributed, multi-node training pipelines in PyTorch and Lightning.
Experience with AWS or another cloud platform, particularly tooling for training and serving medium-to-large models.
Strong software engineering fundamentals, including production-quality Python and a focus on reliability, observability, and cost.
Fluency with modern agentic coding tools such as Claude Code, Cursor, or Codex, and current knowledge of how to use them effectively.

Nice To Haves

Experience training large diffusion or transformer models, ideally on video or time-series data.
Experience with multimodal or spatiotemporal data, geospatial data, or geoscience problems.
Experience serving models in production and integrating them with downstream applications or agents.
Startup experience and comfort working with a small, fast-moving team.

Responsibilities

Build and operate the ML platform for experiment tracking, distributed training, and inference serving, keeping it fast, reliable, and reproducible.
Work with engineers and scientists to define and maintain interfaces between data and training, data and inference, and inference and the agent.
Build data and model infrastructure for multimodal training across sensors, observation types, and text.
Help scientists test ideas through fast, inexpensive experiments and scale successful approaches to petabyte-scale training runs.
Implement and improve internal and user-facing benchmarks and evaluation frameworks.
Measure and optimize the cost and wall-clock time of experiments, training, and inference.
Work with the product team on how the agent calls the model and consumes its outputs, and build the integration.