GEOSLN

An illustration with public data, not Sea Island's data or systems: a USGS 3DEP elevation model of Buffalo Bayou in Houston, opened as a plain GeoTIFF and as a COG over the same 50 Mbps connection. The timings are measured reads, replayed in real time. Elevation: USGS 3DEP.

Sea Island Software

Three years managing complex geospatial data ingest pipelines, migrating production systems to the cloud, and modernizing legacy code into faster, more stable services.

Converted seven years of production GeoTIFFs, served as GeoServer imagery mosaics, into Cloud Optimized GeoTIFFs stored in S3
Increased resilience with improved AWS CloudFormation templates and auto-scaling rules
Refactored older Python processes onto newer GDAL and cloud-native data formats
Replaced a database GeoJSON assembly step with near-real-time GeoParquet pipelines on Airflow

An illustration with public data, not the client's data or systems: 1,761 public places in Harris County, Texas, screened against FEMA flood zones, shown as the same job on two servers. The 10x is the client's result, not a timing of this map. Flood zones: FEMA National Flood Hazard Layer. Places: © OpenStreetMap contributors. Basemap: Esri and its data providers.

Client GIS processing

10x faster geospatial processing, open source instead of Esri arcpy

A complex spatial workflow ran entirely inside Esri arcpy. It was generally reliable, but slow, manual and hard to scale, with high licensing costs and complex server provisioning. We rebuilt it on an open-source geospatial stack, cut processing time by more than 10x and turned a manual desktop task into one that runs unattended. A workflow that could take days in the cloud now runs in hours.

Moved a proprietary GIS workflow to open-source tools: no licence fees, lower cloud costs and less hands-on work for the client
Designed the cloud architecture and deployed an open-source data pipeline that feeds the client's custom AI image classifier
Built a training process that combines historic and real-time data to produce the right amount of training data, so the model could be sold as an MVP replacing a bulky deterministic process
Delivered all code and setup to the client, who is now building on the MVP

Illustration with public data, not the client's: a real Sentinel-2 search over Harris County, Texas. 228 scenes found, 47 under 1% cloud, the clearest pass downloaded in true colour. Contains modified Copernicus Sentinel data 2025. Basemap: Esri and its data providers.

Satellite imagery for ML

Automated satellite imagery query and acquisition for ML training

Built a stable satellite imagery query and acquisition pipeline that removed the client's main workflow limit, letting them scale and bid on larger contracts. The client runs the code in AWS and connected it to several downstream ML training workflows that fund their business today.

Automated, large-scale imagery search and download
Runs in the client's AWS account
Feeds several downstream ML training workflows

Illustration with public data, not the client's: Fort Bend County, Texas, 2016 to 2025, ending on the land developed since 2016. Data: USDA NASS Cropland Data Layer.

Natural resources

Hard-to-reach USDA cropland data, made decision-ready

Processed the last ten years of several USDA crop classifications into the formats and map views a natural resource client needed for a cropland investment decision. A one-time report that turned previously unreachable data into a digestible form, so the owners could make a large, data-driven decision.

Turned hand-jammed desktop GIS processes into an embedded ArcGIS component the client can keep running on their own
Followed the client's guidelines to clean and filter the data for the specific report they requested
Their data is now properly cleaned, formatted and accessible for their business needs

Illustration with public data, not The Aerospace Corporation's: a Planet Tanager hyperspectral scene of Jacksonville, Florida, swept from 376 to 2,464 nm with the spectra of vegetation, water and buildings. Adapted from Tanager STAC Data, available at www.planet.com/data/stac © 2025 Planet Labs PBC. All Rights Reserved.

The Aerospace Corporation

Delivered enhanced hyperspectral processing tools and web-based apps for imagery discovery

Built a new web-based tool to search, display and query internal satellite imagery, where manual processing had been the only alternative. The imagery became accessible to the whole company. Delivered refactored hyperspectral processing code that ran 2–3× faster.

Refactored hyperspectral processing code to run 2–3× faster
Integrated the processing code into the company-wide codebase, so all future imagery pipelines use the enhanced data processing
Full-stack imagery browser on Django, PostGIS and GeoServer
Made the imagery available company-wide

Clairvoyint AI

From platform vision to production infrastructure

Helping Clairvoyint turn its vision of evidence-grounded analytical agents into scalable cloud infrastructure and production web applications.

Helped turn the platform vision into scalable cloud infrastructure
Set up CI/CD for the platform
Integrated user authentication and authorization across the front-end, back-end, and MCP AI servers
Helped set the boundary between deterministic work and AI-supported work in the app

Listed on Clairvoyint's team page

Geospatial case studies, with measured results

Client work