I design reproducible ETL for large heterogeneous datasets, covering ingestion, conflation, entity resolution, validation, and publication to GeoParquet, PMTiles and vector tiles.
Recent deliveries include CDISAW, a 23.5M-record platform unifying 53 source datasets; AnythingPOI, 22.7M conflated points of interest across six countries; Topodex, a geocoder resolving 894M+ place references across nine gazetteers; and WalkGrid, a routing engine fusing 49 datasets at millisecond response.
I hold a PhD in Geospatial Computer Science, an MSc in Computer Science and a BSc in Information and Communications Technology.
What I am working towards
A place is not the same thing as a coordinate, and the difference is computable. Most spatial infrastructure is built to answer where something is, which leaves the questions people actually ask, about what somewhere is like and whether it is worth going, outside the schema entirely.
My research puts those questions inside it. That has meant modelling environmental preference across 51 features of a city, convening the standards work the field was missing, and proposing a format that lets a language model reason about a place rather than render it. The production side is the same problem at volume: infrastructure only earns trust when somebody who did not build it can pick it up and rely on it.
How I work
Systems at this scale fail quietly. A geocoder resolves the wrong settlement. A join succeeds against the wrong projection and nobody notices for a year. A conflation match looks plausible and is not. So the effort goes into validation, provenance, and making a pipeline legible to whoever inherits it.
The numbers I care most about are the unglamorous ones: a four-hour manual map preparation process cut to minutes, ground truth analysis cut from weeks to days, prototypes taken through to commercial deployment rather than left at the paper.
Technical skills
| Area | Tools |
|---|---|
| Languages | Python (pandas, geopandas, Shapely, PyOsmium), SQL, PySpark, Scala, JavaScript and TypeScript, Node.js |
| Data engineering | ETL and ELT design, Apache Spark, DuckDB, GeoParquet, conflation, entity resolution, validation |
| Spatial data | PostGIS, spatial SQL, H3 discrete global grids, vector tiles, OpenStreetMap, Overture Maps |
| Databases and cloud | PostgreSQL, Elasticsearch, AWS (EC2, S3, Lambda), Docker, FastAPI, REST API design |
| Practice | Git, CI/CD with GitHub Actions, pytest, code review, containerized deployment, orchestration |
| ML and GIS | Embeddings, graph neural networks, NLP geoparsing, LLM integration, QGIS, ArcGIS Pro, FME, MapLibre |
Experience
| Period | Role | Organisation |
|---|---|---|
| 2026 to present | Leverhulme Research Fellow in GeoAI and Data Science | Rights Lab and Leverhulme Centre for Research on Slavery in War, University of Nottingham |
| 2025 to 2026 | Assistant Professor in Computer Science | Birmingham Newman University |
| 2025 to 2026 | Geospatial Developer, Digital Twin Platform (contract) | City as Lab, University of Nottingham |
| 2024 to 2025 | Project Lead, Leisure Walking Systems Working Group | Horizon CDT, University of Nottingham |
| 2020 to 2024 | Doctoral Researcher, Geospatial Data Science | Nottingham Geospatial Institute, Ordnance Survey funded |
Credentials
AWS Certified Cloud Practitioner, 2025 to 2028.
Peer-reviewed publications in platial analysis, leisure walking systems, walkability and spatial data formats, alongside published points of interest datasets covering six countries. The full record, with abstracts, PDFs and citation data, is on the publications page.
Availability
Open to consulting engagements and research collaborations in geospatial data engineering, spatial platforms and GeoAI. The fastest route is email, at james@jameswil.com.
