Public geospatial sources
NYC/NYS building footprints, OSM roads, footways, crossings, POIs, open space, and water context.
QGIS portfolio / AI data QA
QGIS Portfolio
I built a QGIS asset production package for Lower Manhattan that converts public geospatial sources into normalized, QA-checked, AI-reviewable GIS layers.
7,219
normalized GIS features
10
asset layers
0
invalid geometries after QA
EPSG:2263
NYC working CRS
Project set
QGIS assets first, evaluation toolkit second.
The core proof is the Lower Manhattan QGIS handoff package. The evaluation lab pairs with it for AI reviewer workflows; RPInSight is only supporting evidence of geospatial product work.
Flagship 1
AI-Ready QGIS Assets
Lower Manhattan QGIS package with reviewed layers, QA flags, GeoPackage exports, docs, and screenshots.
7,219 features
10 layers
0 invalid geometries
Flagship 2
GIS AI Evaluation Lab
QGIS-ready evaluation project for grading AI answers to CRS, topology, exports, licensing, and review tasks.
25 tasks
101 concepts
QGIS .qgz
Adjacent product
RPInSight
Campus map assistant showing the same geospatial QA instincts inside a user-facing Mapbox product.
Mapbox
AI search
Campus data
Evidence package
Everything a reviewer needs to inspect the work.
The package includes a QGIS project, GeoPackage exports, GeoJSON exports, a data dictionary, QA report, labeling guide, source notes, and AI evaluator task bank.
Technical handoff
Plain asset inventory.
| Asset | Format | Purpose |
|---|---|---|
| Lower Manhattan GIS Package | .gpkg | Primary multi-layer spatial dataset |
| QGIS Project | .qgz | Styled QGIS review workspace |
| GeoJSON Exports | .geojson | Web/export-friendly layer outputs |
| QA Report | .md | Geometry, CRS, confidence, and review checks |
| Data Dictionary | .md | Field schema and allowed values |
| Labeling Guide | .md | Human review protocol |
| Source Notes | .md | Source, license, and limitation notes |
| AI Trainer Task Bank | .md | GIS/QGIS evaluation prompts |
Asset production pipeline
Clean spatial inputs into AI-ready GIS assets.
The project normalizes, clips, validates, flags, reviews, and packages public geospatial data into a compact multi-layer handoff that can be inspected in QGIS or consumed downstream as GeoPackage and GeoJSON.
Normalize, clip, validate
All features are clipped to the Lower Manhattan ROI and normalized to one review-friendly schema.
Export-ready deliverables
QGIS project, GeoPackage, GeoJSON exports, QA report, labeling guide, and AI trainer task bank.
QGIS desktop proof
The assets open as a real QGIS review workspace.
Screenshots captured directly from the QGIS desktop application on macOS show the layer panel, styled map canvas, project CRS, and attribute table.
QA surface
Ambiguity is preserved instead of hidden.
Dense urban GIS data is messy. The portfolio includes `confidence`, `review_status`, `qa_flag`, and a spatial `qa_issues` layer so a reviewer can see exactly where the dataset should be trusted, checked, or treated as approximate.
Schema discipline
Every feature carries provenance and review state.
The core schema is intentionally simple: asset id, type, label, source, source id, confidence, review status, QA flag, notes, and update date. That makes the data legible for AI evaluation workflows, not just map display.
Layer inventory
Built like a handoff package.
AI trainer readiness
Includes the questions a GIS evaluator would ask.
The GitHub documentation includes a task bank with QGIS-specific prompts and answer keys covering CRS choice, geometry validation, crosswalk ambiguity, topology checks, exports, and AI response review.