Hi! I'm Luke. I architect and build ML systems that solve complex geospatial challenges at scale. With over a decade of experience spanning both startup and research environments, I've led the development of AI solutions for environmental monitoring, real estate analytics, and land management. I specialize in designing robust ML pipelines and spatially-aware data systems that transform raw geospatial data into actionable insights.
Resume
My career has evolved from academic research to building production ML systems at scale. At Vibrant Planet, I lead several ML development efforts for a cutting-edge geospatial platform, designing and deploying pipelines that process massive datasets for critical decision-making. Previously, as Head of Data Science at Iggy, I led the development of novel geospatial data products for partners in real estate and private equity. I've consistently worked at the intersection of machine learning, spatial data, and scalable systems.
Learn more about Vibrant Planet
I specialize in building end-to-end ML systems that handle complex geospatial data at scale, with particular expertise in computer vision, remote sensing, and distributed computing.
PyTorch, ONNX, Kubeflow, Ray
AWS, GCP, Azure, Airflow
GDAL/OGR, PostGIS, Earth Engine
Python, R, SQL
A few highlights from some of past and ongoing work
Developing a state-of-the-art live aboveground biomass model at 10m resolution. This model leverages a custom Feature Pyramid Network, powered by features extracted Vibrant Planet's foundation model (a Masked AutoEncoder). Input data includes Sentinel-2 imagery, terrain features, and climate variables, enabling accurate and high-resolution biomass and carbon estimates.
Using a host of ML approaches (Bayesian hierarchical allometric models and computer vision) to assign critical structural attributes to individual trees, including size/volume and status (alive or dead) across the western US.
Insights about the demography and amenities (within walking or driving distance) at any address in the US.
A data diligence process, including an interactive mapping application and form to systematically and iteratively validate and improve Iggy's features. The process included a number of aggregate checks (statistical measures, and spatial distributions or spatial consistency of individual features) and spot checks (on individual data points, including ID, location, value).
The first comprehensive analysis of human impacts on lands in the American West
Learn more
Bayesian hierarchical models for complex monitoring data
Learn more
Artificial intelligence to predict Amazonian river health
Find the code here
Tools for integrating multi-scale earth observation data streams
Learn more