Custom data
“We need 5M structured vector examples.”
The catalogue is the starting point, not the limit. The team and network of creators who produced it also build datasets to a written specification.
Workflow
| Step | Stage | What happens |
|---|---|---|
| 01 | Define requirements | Task, modality, scale, formats, target metrics. |
| 02 | Define taxonomy | Sectors, subjects, chart families, room types, page formats. |
| 03 | Source or commission | From the catalogue, or created for the brief. |
| 04 | Clean | Normalise, strip raster, reconcile fonts, remove empty groups. |
| 05 | Validate | Parse, render, compare renderers, check dimensions. |
| 06 | Deduplicate | Structural hash and perceptual hash with published thresholds. |
| 07 | Metadata | Full record schema, captions, layout boxes, text roles. |
| 08 | Quality score | Component scores and badges for the delivered set. |
| 09 | Sample approval | You sign off on a stratified sample before the full run. |
| 10 | Enterprise licence | Exclusivity, term and territory as agreed. |
| 11 | Delivery | Scoped storage, checksums, Croissant manifest. |
| 12 | Versioning | Semver releases with change notices. |
Services
- Custom SVG dataset to a brief
- Industry-specific corpus
- Geographic or language-specific corpus
- Architectural corpus with custom unit typologies
- Design corpus with constrained palettes, formats or brand rules
- Lineage tagged per asset, so every file in the delivery can be traced
- Evaluation dataset with held-out stratification
Provenance
Assets are composed, created and processed into SVG from source files by The Doodle Desk's team and network of creators. Every figure, name and caption is a written placeholder; no real organisation, person or measurement appears.
What we will not do
Deliver files we cannot document per asset.