Architectural RDM-Pipeline

Architectural RDM-Pipeline is a browser-based tool for turning tabular research data into structured RDF datasets. It combines SHACL application profiles, CSV or Airtable sources, visual mapping, enrichment services, SHACL validation, metadata capture, and RO-Crate publishing in one Vue single-page application.

The app is designed for research data workflows where datasets should be understandable, reproducible, and publishable without hand-writing Turtle files or maintaining one-off conversion scripts.

Demo

What You Can Do

• Load SHACL profiles in Turtle format and resolve owl:imports.
• Prepare source data from CSV files or Airtable tables.
• Load a built-in showcase project from the mapping view for demonstrations or onboarding.
• Build mappings visually on a Vue Flow canvas by connecting source columns to SHACL property shapes.
• Choose a subject identifier column for each mapped shape and original table. One table can supply Building, Location, and Stakeholder profiles with separate identifiers, while declared sh:node relationships keep them linked.
• Detect and display table-to-table references from Airtable-style linked record fields.
• Add enrichment nodes for GeoNames and Lobid/GND workflows. Choose the GeoNames input mode from the dropdown on the node or in its configuration: GeoNames ID, free text, or latitude/longitude for the nearest city or populated place. The canvas displays one input for ID/text and two inputs for latitude/longitude. Coordinate fields must come from the same table and use decimal degrees. The coordinate modes use the GeoNames reverse geocoding service. New nodes select Name, ID, Latitude, Longitude, and Country name by default. Other output fields remain available but are initially unselected. Changing mode clears previous results and removes incompatible input connections. Free text uses the first result from the GeoNames search service. Include a country or region for more specific searches. Failed lookups leave their output rows empty while the remaining locations are processed. The node and its configuration panel report each failed input and its error message.
• Transform latitude/longitude columns into GeoSPARQL WKT points.
• Combine two or more columns from the same table with the merge transformation. Use Add input for additional fields. Values are joined in input order with the editable Separator, which defaults to , . An empty separator concatenates values directly. Empty values are skipped. Arrays from Airtable contribute their individual elements using the same separator. Connect the merged output to a SHACL property or the GeoNames free text input, for example to search for Berlin, Germany from separate city and country columns. Merge also accepts fields from the same GeoNames node, for example its Name and Country name outputs. Run GeoNames to populate or refresh the merged output. Merge also accepts fields from one upstream split or merge node. Connections that would create a cycle are rejected.
• Split a field using the split transformation and an editable literal separator (default , ). Choose Two columns to split at the first separator into First part and Remaining text. Both columns keep the original row count and record IDs: missing parts remain empty, and additional text stays in the second column. For example, Paris, France, Europe becomes Paris and France, Europe. Choose Array to split at every separator into one array cell per row. Existing Airtable arrays keep their elements; a comma inside an array element is not split again. Arrays of objects need a text or ID field extracted first. Split accepts table, GeoNames, merge, or split outputs. Its output columns can feed SHACL properties, merge, another split, or the GeoNames inputs. Changing output mode removes incompatible connections. Settings are saved with projects. Re-run GeoNames after changing its transformation inputs to fetch new lookups. For multiple values on separate lines inside one CSV cell, choose Array and the Line break separator preset (or enter \n). Space selects one ordinary space; line breaks and spaces are distinct separators. The separator field also supports \t for tabs, \r\n for CRLF, and \\ for a literal backslash. A visible label distinguishes a space from an empty separator.
• Create reproducible identifiers with the HASH transformation. Connect a name field (for example XY Architects) to generate its SHA-256 hash as 64 lowercase hexadecimal characters. Surrounding whitespace is trimmed; case and internal spaces are preserved. Empty values stay empty, and arrays are hashed element by element without changing row IDs or element positions. HASH accepts table, GeoNames, merge, split, or other HASH outputs. Its output can feed properties, transformations, or a sh:node reference. Connecting it to Building's Architect reference automatically makes the hash the Architect identifier while Building keeps its original ID; map the original name to Architect's name property. You can also select the HASH output's hash field in a shape's Subject identifier dropdown. The node and connections are saved with the project.
• Generate RDF from the current mapping.
• Review generated RDF subjects as cards, tables, or Turtle.
• Validate generated RDF against the loaded SHACL profiles.
• Capture RO-Crate dataset metadata through a SHACL-form based publish view.
• Publish a ZIP package containing:
  • ro-crate-metadata.json
  • generated RDF as data/dataset.ttl
  • source tables as CSV
  • materialized enrichment or transform outputs as CSV when they are part of the exportable pipeline state
  • resolved SHACL profiles
  • RML mapping output

Array cells are shown as JSON arrays in previews and archived source CSVs. Mapped arrays generate repeated RDF property values rather than ordered RDF collections. RML uses additional scalar CSV rows that retain the resolved entity URLs, so individual array values (including values containing commas) are preserved.

Who It Is For

This tool is aimed at research data managers, data stewards, and domain experts who need to prepare RDF datasets from tabular data while staying aligned with SHACL-based application profiles. It is especially useful when a dataset needs both machine-readable RDF and packaging metadata for publication or archival handover.

Main Workflow

1. Open the Prepare view.
2. Optionally load the built-in showcase project from Options for a ready-made example.
3. Add source data, for example a CSV file or an Airtable base.
4. Add a target schema by uploading a SHACL profile or loading an embedded profile.
5. Connect table columns to SHACL properties on the canvas.
6. Optionally add enrichment or transformation nodes.
7. Review generated RDF in the Review view.
8. Check validation results.
9. Complete dataset metadata in the Publish view.
10. Publish the RO-Crate ZIP.

Entity identifiers and relationships

After mapping a column to a shape, its canvas node shows a Subject identifier dropdown for each original table. The dropdown groups columns from that table and its enrichment outputs. Select the CSV's building_id on Building and the connect the GeoNames output's id to Building's Location (sh:node) property. For profiles fed from the same original table, the referenced shape inherits the connected field as its subject identifier. Location shows that field and Inherited from Building → Location, instead of a separate selector. This also works with a CSV column directly: a geonames_id column connected to the reference becomes Location's identifier. Building keeps its CSV identifier. When GeoNames has not run yet, the inheritance takes effect as soon as its output is available. Changing the reference connection updates the inherited identifier; removing it restores the independent selector. Saved connections carry this rule without storing another primary-key selection on the target shape. With target classes ex:Building and ex:Location, the generated subjects are http://example.org/Building/B1 and http://example.org/Location/2950159. The selected value supplies the URL suffix; the target class supplies the class segment. Without a target class, the shape name supplies that segment. Default keeps the existing storage record ID (or first column for a source without record IDs) from the original table. Selection changes RDF identity without changing source row IDs. All mapped properties for the same profile and original row use this identity, including properties supplied by enrichment outputs. Connecting a GeoNames ID to Building's Location reference therefore adds a link on the existing Building; it does not create a Building whose ID is the GeoNames ID. Multiple Buildings keep their individual CSV IDs while sharing one Location.

If Building declares a property with sh:node ex:LocationShape, and both shapes receive data from the same row, ARDMP automatically links the Building to that row's Location. This also works through GeoNames, Lobid, and merge outputs that retain their source lineage and row IDs. A reference column explicitly mapped to the property takes precedence and defines the target identity for the same original table. For unrelated tables, references can contain the target's selected ID or an Airtable storage record ID; ARDMP resolves either to the target entity URL while preserving that table's independently selected identifier.

An array connected to a sh:node reference also supplies the referenced shape's identifiers: each nonempty value creates a separate entity linked from the original row. For example, split a multiline CSV cell into an array, connect it to Building's Source reference and Source's URL property. Each URL identifies one Source with its own URL, while Building keeps its original ID. When a shape uses an array identifier, other array fields align by position; scalar fields apply to each entity. Unfilled positions stay empty.

Selected identifiers are trimmed and URL-encoded. Repeated IDs share a subject, so several buildings can reference one Location or Source. Empty identifiers, objects and nested arrays are skipped; a row with no valid identifier creates no entities or automatic incoming links. The canvas reports the affected row count. A missing selected column requires choosing another column. The project saves these choices. RO-Crate exports include generated entity CSVs with resolved subject and reference URLs, used by the RML mapping to preserve the identities and links shown in the RDF preview.

Technology

• Vue 3 and TypeScript
• Vite
• Pinia
• Vue Router
• PrimeVue
• Vue Flow and Dagre
• rdflib
• shacl-engine
• @ulb-darmstadt/shacl-form
• PapaParse
• localForage
• JSZip
• Vitest

Development

Recommended runtime: Node.js 20 or newer.

npm install
npm run dev
npm run type-check
npm run lint
npm run test
npm run build

The local dev server is served by Vite, usually at:

http://localhost:5173/

Repository Layout

src/
  assets/profiles/      bundled SHACL profiles
  components/           app-wide shared UI
  domain/               framework-light domain models
  features/             feature modules for mapping, browse, export, SHACL UI
  router/               route definitions
  services/             RDF, validation, export, mapping, project, infrastructure
  stores/               Pinia stores
  styles/               global SCSS and design tokens
  views/                route-level views

Deployment

The app is a static hash-routed SPA and can be deployed to GitHub Pages. The production base path is derived in vite.config.ts from the repository name when the GitHub Actions workflow runs, so the source code does not hardcode a Pages path.

The deployment workflow:

• installs dependencies with npm ci
• runs type checking
• builds the Vite app
• uploads dist/
• deploys through GitHub Pages

For GitHub Pages, set the repository Pages source to GitHub Actions.

Project Status

The project is under active development. The current architecture separates domain models, Pinia state, feature modules, infrastructure adapters, RDF generation, validation, and export logic. Mapping extensions are being moved toward explicit module boundaries so new importers, enrichers, transformations, and exporters can be added without growing route views or central stores.

License

MIT. See LICENSE.