Skip to main content

Create an AI Agent-Ready Dataset

FactVerse AI Agent workflows need data that is traceable, fresh enough for the task, mapped to the right operational identity, and reviewed by the right owner. Use this recipe to prepare a DFS Pro dataset for AI Agent workflows.

Outcome​

At the end of this workflow, you should have:

  • a defined Agent question or workflow;
  • source systems connected or imported;
  • mappings reviewed;
  • a DFS Pro dataset created;
  • profile and quality issues reviewed;
  • steward assigned;
  • dataset validated;
  • source limitations documented for Agent users.

Step 1: Define the Agent workflow​

Start with the operational task, not the data table.

Examples:

  • summarize equipment status before inspection;
  • prepare predictive maintenance evidence;
  • compare recent alarms with work orders;
  • explain why an energy or facility value changed;
  • prepare a reviewed work-order draft.

Record:

  • user role;
  • site or tenant boundary;
  • assets or systems involved;
  • expected output;
  • reviewer;
  • allowed action boundary.

Step 2: Identify required evidence​

List the evidence the Agent should use.

Evidence typeExamples
Asset contextAsset ID, equipment name, aliases, location, system relationship.
SignalsSensor values, meter values, status, event history.
Work recordsWork orders, inspection records, close notes, operator actions.
DocumentsSOP, maintenance manual, drawings, compliance notes.
Quality notesMissing values, stale values, source limitations.

Step 3: Connect or import source data​

Use DFS Lite connectors for live or scheduled sources. Use DFS Pro dataset creation for imported or prepared data.

For each source, confirm:

  • source owner;
  • update frequency;
  • timestamp field;
  • identity field;
  • unit and value range;
  • permission to use the data in the Agent workflow.

Step 4: Map operational identity​

AI Agent output is only useful when identity is stable.

Review:

  • source equipment names;
  • asset IDs;
  • aliases;
  • location hierarchy;
  • point names;
  • work-order asset references;
  • document references.

If source systems use different names for the same asset, create or update mappings before validating the dataset.

Step 5: Create DFS Pro dataset​

In Dataset Center:

  1. Create dataset.
  2. Choose source type.
  3. Add name and description.
  4. Add or confirm source ID.
  5. Save.
  6. Open detail page.
  7. Preview rows.
  8. Review profile.

Dataset description should include:

  • source system;
  • purpose;
  • owner;
  • expected refresh cadence;
  • known limitations;
  • intended Agent workflow.

Step 6: Review quality and profile​

Check:

  • row count;
  • required columns;
  • null ratio;
  • distinct IDs;
  • timestamp range;
  • stale records;
  • quality issues;
  • schema changes.

For Agent workflows, document limitations in plain language. Example: "Work-order close notes are available from the CMMS feed, but part replacement details are only available after nightly sync."

Step 7: Assign steward and validate​

Assign a steward who can approve the dataset for the Agent workflow.

Validate when:

  • identity mapping is reviewed;
  • source timestamps are understood;
  • quality issues are acceptable for the task;
  • dataset schema is stable enough;
  • downstream reviewer accepts the assumptions.

When using the dataset in an Agent workflow, keep these visible:

  • dataset name and version;
  • source timestamp;
  • quality notes;
  • reviewer;
  • allowed output type;
  • action approval requirement.

For MCP workflows, confirm scopes and endpoint access separately in the MCP documentation.

PageUse
DFS Pro DatasetsCreate and validate the dataset.
Data QualityReview source freshness and quality.
FactVerse AI Agent Data ReadinessCheck whether source evidence is ready for Agent use.
MCP Scope MatrixPlan endpoint and scope access for MCP clients.