AI Research Scientist

hum.ai•San Francisco, CA

About The Position

Hum is building a world model for Earth that represents the dynamics of Earth processes, forecasts how places will change, and makes that understanding useful. The AI Research Scientist will own the model at the center of the system. The engineering team is building pipelines to ingest petabytes of Earth observation and remote sensing data, as well as an agent that uses the data and models to answer customer questions. The scientist will work closely with the Chief Scientist to decide what the model learns and how, and with engineers to train, experiment with, and deploy models for customers and broader impact. The role involves evaluating and evolving the model architecture, interpreting large and unusual datasets with domain experts, defining measures of success, and developing ideas from small, resource-efficient experiments through to petabyte-scale training. The scientist will shape how the model integrates with the agent and product, help set milestones, roadmaps, and resourcing with leadership, identify risks, and publish research on Earth observation foundation models while representing the work at conferences and in the broader community.

Requirements

PhD in computer science, engineering, a related field, or equivalent experience.
At least 2 years of industry experience building and shipping AI systems.
Deep understanding of cutting-edge AI research, with a record of original ideas or meaningful contributions demonstrated through publications and/or projects.
Experience taking ideas from small, exploratory experiments in compute- and data-limited settings to performant, large-scale training runs.
Comfort designing models for unconventional data modalities beyond text.

Nice To Haves

Experience training models on non-language data such as images, time series, or sensor data, ideally geospatial or Earth observation data, or a background in Earth science.
Background in mathematics or physics, especially where it informs modeling physical systems and sensor data.
Strong engineering fundamentals, including fluency in Python and PyTorch, ability to write efficient and reliable training and inference code, and comfort using agentic coding tools.

Responsibilities

Own and evaluate the model architecture, changing the approach when evidence supports doing so.
Apply state-of-the-art techniques to uncover semantics in large, unusual datasets and translate findings into downstream performance, working with domain experts.
Help define measures of success rather than relying on existing GeoAI benchmarks.
Test ideas through fast, low-cost experiments and scale successful approaches to petabyte-scale training with engineers.
Work with the agent team to determine how the model is called, what it returns, and how it can be more useful in the product.
Set milestones, roadmaps, and resourcing with leadership, and identify risks early.
Write and publish research on Earth observation foundation models and represent the work at conferences and in the broader community.