Workflow Editor - Entity Enricher Documentation

Workflow Editor

The page where you build one enrichment end to end — sample data, schema, models, results. At its centre sits the schema tree: a visual editor with drag-and-drop reordering, inline editing, keyboard navigation, and AI-assisted modifications.

Overview

The schema tree lets you view and modify your enrichment schemas without writing JSON by hand. Properties are displayed as an interactive tree grouped by expertise domain, so you can see at a glance which fields belong to which specialist area. Every change is tracked with undo/redo history and auto-saved after a 2-second debounce.

Key Capabilities

Expertise Grouping

Properties grouped by expertise domain with collapsible sections. Toggle grouping on/off.

Drag-and-Drop

With the Sorted view off, reorder properties within or across groups by dragging — the arrangement is saved on the schema. Visual drop target indicators show where items will land.

Inline Editing

Click any property to edit its name, type, description, and examples directly in the tree.

Keyboard Navigation

Arrow keys to navigate and to collapse or expand, Enter to edit, Delete to remove, Ctrl+A to select all. Undo and redo are toolbar buttons, and the history survives switching schemas or reloading the page.

Search & Filter

Text search to find fields in large schemas: property names first, falling back to descriptions and examples when no name matches. Nested matches are revealed with only the path leading to them.

Undo / Redo

Full history stack for manual edits. AI-generated changes replace without adding undo entries.

Types & Value Sets

Tabs across the top hold the schema's named types. The + button creates either an Object (a nested entity with its own properties) or an Enum (a closed list of allowed values you can point any text property at).

Diagram View

The DIAGRAM view next to SCHEMA and JSON draws the schema as an entity diagram — named entities as titled cards, nested objects as plain cards named by their relation, with references (1–1) and links (1–N) as dashed edges. It follows your edits live, and clicking an entity card opens it in the SCHEMA view.

The Property Tree

Each row in the editor represents a schema property with its key attributes visible at a glance:

SelProperty NameTypeDescription
□company_name
🔍
stringLegal company name
□revenuenumberAnnual revenue in USD
□headquartersobjectCorporate headquarters
□ citystringMunicipality name
□ country
🔍
stringNation where HQ is located
□gmp_statusbooleanGMP certification status

Nested properties (like city under headquarters) are indented to show hierarchy. The magnifying glass icon marks search keys — fields used to identify the entity.

Value Sets (Enums)

A value set restricts a text property to a fixed list of allowed values — a status, a grade, a classification code. Enrichment then cannot return anything outside that list, which keeps the same concept from arriving as three different spellings across runs.

Working with value sets

Create one

Press + on the type tabs, choose Enum, and name it. The new tab opens its member list.

Edit members

Add, rename, reorder or remove values. Give a member a note when its meaning is not obvious from its name.

Use it

Set any text property's Type to the value set (or to an array of it) from the Value Sets group in the type dropdown.

Undo it

Set the property's Type back to string, or delete the value set — properties using it become ordinary text properties again.

Only use a value set for a vocabulary you can list completely. If a real entity needs a value you left out, the model is forced to pick the closest wrong one instead of reporting the problem — so an open-ended field (a name, a description, a free-text category) should stay plain text. Schema generation proposes value sets conservatively for the same reason, and everything it proposes is editable here.

Editing Properties

Click a property row or press Enter to switch to edit mode. In edit mode, you can modify:

NameThe property key in the JSON output. Use snake_case for consistency.
TypeSelect from string, number, integer, boolean, array, or object. Changing type updates the enrichment output format.
DescriptionContextual description that guides the AI during enrichment. More detail leads to better results.
ExamplesTwo or three realistic example values. These help the AI understand the expected format and content.
ExpertiseWhich domain expert provides this value. Determines grouping in multi-expertise enrichment.
NullableWhether a real instance can genuinely lack this value. Non-nullable fields are required for database admission — an enrichment that cannot fill them is rejected by the quality gate. Unavailable on a database key (a null value identifies no row) and on a preserved field (your input always supplies it).
MultilingualWhether the field should be enriched in multiple languages.
PreserveWhether to keep the original input value unchanged. For values your own system owns — a database id, an internal SKU, a customer reference — not for public identifiers (CAS, ISBN, EAN, a ticker) the enrichment can look up for you. Every enrichment must then provide a value for it.

Keyboard Shortcuts

The schema editor supports a full keyboard-driven workflow for efficient editing:

ShortcutAction
Arrow Up / DownNavigate between properties (hold Shift to extend the selection)
Arrow Left / RightCollapse or expand the selected object
Home / EndJump to the first or last property
EnterToggle edit mode on selected property
Delete / BackspaceRemove selected property
Ctrl+A / Cmd+ASelect every property
EscapeExit edit mode or clear selection

Drag-and-Drop Reordering

Reordering by hand requires turning the Sorted toggle off. The editor lists properties alphabetically by default, and dragging is disabled there — a hand-made order that a view setting could silently override would not be worth much. Unchecking Sorted switches to the schema's own arrangement and enables dragging; a visual insertion line shows exactly where the property will land, and the resulting order is saved with the schema. You can drag within an expertise group or across groups to change the domain assignment.

When expertise grouping is enabled, dropping a property into a different group automatically updates its expertise assignment. This makes reorganizing large schemas intuitive — just drag a field from “Financial Analyst” to “Regulatory Expert” and its expertise updates accordingly.

AI Edit Pane

Below the property tree, the AI edit pane lets you modify the schema using natural language. Type an instruction like “add a parent_company reference” and the AI applies the change while preserving your existing structure.

AI Suggestions

After each generation or edit, the AI provides 5 improvement suggestions displayed as clickable chips. Click a suggestion to automatically fill the edit input and apply it. Suggestions cover data completeness, quality, relationships, internationalization, and domain-specific enhancements.

Learn more about AI schema editing in the AI Schema Generation guide.

Ambiguity Badges

Read a property's name in the context of its parent object: how many distinct things could it be asking for? Exactly one is clear. Two or more, and the property carries an ambiguous badge — amber when the readings mostly overlap and only edge cases differ, red when they would yield materially different data. Hover the badge to read the competing interpretations the name admits, and the rewritten description that pins exactly one of them. A name the parent has nothing to answer is flagged too: there the tooltip lists the readings that were considered and rejected, and the fix is a rename or a removal — no description can give an entity a property it does not have.

Re-check button

The shield icon in the toolbar analyzes properties that have no annotation yet, and switches to a full re-analysis once everything is annotated. Each run is a small billed AI call.

The remedy is a description here

On a saved schema the badge suggests a rewritten description, never a rename: the schema is live, so pinning the meaning in the description leaves the data contract intact. A description that only restates the name adds nothing, and the property is judged as if it had none.

Stale annotations

Renaming a property or editing its description drops its annotation — the verdict was rendered on both. The editor highlights these properties and offers a re-check.

Per-schema toggle

Disable the check for a schema from the overflow menu — badges and warnings disappear, and re-enabling on a never-analyzed schema runs the check automatically.

The full mechanics — the two verdicts, the interpretations behind each one, and why the fix is a rename before the schema exists but a description once it is live — are covered in the Ambiguity Check guide.

Schema Management

The schema selector lets you create, load, pin, duplicate, and delete schemas. Schemas are saved per organization and accessible to all team members.

Auto-Save

Changes are saved automatically after a 2-second pause, with content-hash deduplication to avoid unnecessary writes.

Pin Schemas

Pin frequently used schemas to keep them at the top of the selector list.

Duplicate

Create a copy of any schema to experiment with changes without affecting the original.

Soft Delete

Deleted schemas can be restored. Permanent deletion is a separate action.

Next Steps