Labs · show, don't tell

Live demos,
real data.

We'd rather show you a working system than a screenshot. Every demo below runs live in your browser on real open or client data — click in and explore. Grouped by what they prove.

Imagery & computer vision

Trained models on real imagery — with honest, held-out accuracy numbers, not marketing claims.

CV + HITL · ocean plastic

Marine Debris Detector

A real U-Net trained on the MARIDA Sentinel-2 archive spots floating plastic in ocean scenes. Toggle the model, the FDI spectral index and ground truth; review, accept, reject and relabel detections — and watch precision/recall update live against the truth.

Semantic segmentationMLOps & model validation
Deep learning · forestry

Tree-Crown Delineation

A DeepForest deep-learning detector plus NDVI/GrabCut segmentation turns a NAIP orthophoto into QA-ready individual tree-crown polygons — with a real hand-checked precision number, tree count, canopy area, and GeoJSON / GeoPackage export.

Computer vision & object detectionRemote sensing
CV + HITL · deep-sea imagery

Abyssal Vision

A deep-sea dive line, end to end: a Google-Maps-style seafloor mosaic with automated polymetallic-nodule detection and real-unit sizing, a megafauna expert-review queue, and a live analytics dashboard — de-duplicated across the 70% frame overlap, sized from the georeferencing.

Computer vision & object detectionMLOps & model validation
ML + remote sensing · mineral exploration

Gold Prospectivity — Mauritania

A machine-learning mineral prospectivity map of the Tasiast–Tijirit greenstone belt: airborne geophysics, Sentinel-2 and ASTER alteration mapping, structural lineaments and geology fused on one 250 m grid — validated by a spatial holdout test in which the withheld Tasiast mine district ranked in the 99th percentile. Click any cell for its evidence drivers.

Remote sensing & geophysicsMLOps & model validation

Real-time & environmental

Live feeds and physics — air, water and weather, streamed into decisions as they change.

Property, site selection & land

Parcels, zoning, constraints and measurement — the questions real-estate and land decisions actually turn on.

Cartography, precision mapping & publication

Print- and journal-ready output, and survey-grade vector geometry — automated, reproducible, exact.

Data pipelines & automation

Ingest, scrape and structure — the plumbing that turns messy public sources into decision-ready data.

Every demo runs on real data. The Marine Debris Detector runs a real U-Net (trained by GIA on the open MARIDA Sentinel-2 archive) on bundled patches, with optional live in-browser ONNX inference; Tree-Crown Delineation runs a real DeepForest model with a hand-checked precision figure. AirGrid and the Weather Radar run on live open APIs (Open-Meteo, RainViewer). Parcel Screening, Site Due-Diligence and Property Planning run on real open government data (Levy County FL, Athens-Clarke GA + FEMA + USGS, and Vicmap AU). Spain Climate, City Posters and the KTEB airport map are reproducible cartography from open sources (Köppen-Geiger, OpenStreetMap, FAA NASR). Abyssal Vision is a bounded pilot on one sample seafloor dive line — nodules detected and sized by the pipeline, megafauna routed to expert review. New demos are added as we build them.