Strength

AI-⁠assisted ETL for spatial data

Get from legacy data to a trusted spatial database in weeks, not months, and know exactly what happened to every record.

Talk to us
The reality

Data work is where GIS projects lose time

Network registries, cadastral extracts, survey files, CAD drawings, spreadsheets, scanned plans. Different coordinate systems, inconsistent attributes, duplicated and broken geometries, and documentation that is missing or wrong.

Writing the mappings and checks for that by hand takes months. Large language models are good at exactly this kind of tedious, pattern-heavy work, if you keep them on a short leash.

Where AI helps

Five steps, accelerated

  • 1

    Profile

    Summarise unknown sources: attributes, value patterns, geometry types, coordinate systems, outliers and likely keys.

  • 2

    Map

    Propose how source fields map to the target data model, including code lists and free-text normalisation, with reasons.

  • 3

    Generate

    Draft the transformation code: GDAL/OGR, SQL for PostGIS, Python, orchestrated by a standard workflow tool.

  • 4

    Validate

    Write and run quality rules: geometry validity, topology, domain checks, duplicates, plausibility. Flag suspicious records for review.

  • 5

    Document

    Generate data dictionaries, lineage and a reconciliation report: how many records came in, went out and were rejected, and why.

Our rules

AI accelerates, engineers decide

  • Human review: every mapping and rule is reviewed by a GIS engineer before it is used
  • Deterministic output: the pipeline that runs in production is ordinary, versioned code, not a model improvising on your data
  • Full traceability: every source record can be followed to its target or to a documented rejection
  • Your data stays yours: for sensitive or government data we use locally hosted open-weight models, with no data leaving your environment
  • Repeatable: the same pipeline re-runs on the next delivery of data
Typical uses

Where it pays off most

Got a pile of data nobody trusts?

Send us a sample. We will show you what the pipeline would look like.

Send us a sample