1. Ema v2 Docs
  2. Templates
  3. Data Extraction Pipeline AI Employee

Data Extraction Pipeline AI Employee

The Data Extraction Pipeline AI Employee is built from the Data Extraction Pipeline curated template (category: IT & Ops). It is a form AI Employee (AIE): you submit a document, and it extracts structured data according to a configurable schema. It is the general-purpose extraction starting point — useful for invoices, contracts, forms, receipts, or any document with a repeatable structure. It automates manual data entry, processes documents in seconds, and scales to volume spikes without additional staff.

Workflow

The template ships a single-agent workflow:

NodeAgent typeWhat it does
Extract DataextractionExtracts structured data from the document according to the configured output schema, returning the exact value for each field and flagging any field that is ambiguous or missing.

A start node feeds the document into the graph, and a publish node exposes the extracted data. The edges run start → extract → publish.

The Extract Data node ships with a general-purpose output schema in its type_config, which you replace with the fields you need. The default schema captures common document attributes:

{
  "output_schema": {
    "type": "object",
    "properties": {
      "document_type": { "type": "string", "description": "Type of document (invoice, contract, receipt, report, etc.)" },
      "document_date": { "type": "string", "description": "Date on the document" },
      "parties":       { "type": "array", "items": { "type": "string" }, "description": "Names of parties or organizations mentioned" },
      "key_values":    { "type": "array", "items": { "type": "object",
                          "properties": { "field": { "type": "string" }, "value": { "type": "string" } } },
                         "description": "Key field-value pairs extracted from the document" },
      "amounts":       { "type": "array", "items": { "type": "object",
                          "properties": { "label": { "type": "string" }, "value": { "type": "number" }, "currency": { "type": "string" } } },
                         "description": "Monetary amounts found in the document" },
      "summary":       { "type": "string", "description": "Brief summary of the document content" }
    },
    "required": ["document_type", "key_values"]
  }
}

Starting configuration

The template clones in with this configuration, which you can change in the builder:

  • Model selection — Let Fusion Pick (EmaFusion™ chooses the model per request).
  • Optimization priority — Most accurate.
  • Conversation context — disabled (this is a single-submission form workflow, not a chat).
  • Feedback — enabled, with thumbs-up / thumbs-down reasons and free-text comments.

This template does not ship with a recommended-data-sources note, because it works directly on the documents you submit rather than on a connected Knowledge base. Where the documents you process reference internal terms or codes, you can attach a Knowledge base to improve interpretation.

How to use

After you select Use template and the AI Employee opens in the builder:

  1. Define your extraction schema. Edit the output schema in the Extract Data node with your field names, types, and descriptions.
  2. Test and refine. Submit sample documents to check extraction accuracy and tighten the schema and Instructions where the agent is uncertain.
  3. Connect output Tools. Add Tools to push the extracted data to your downstream systems — a database, spreadsheet, or API. See Tools.
  4. Publish and submit. Publish the workflow, then submit documents and receive structured output.

If your documents are a specific, well-known type, start from a purpose-built template instead — Invoice Extraction and KYC Document Review ship with tuned schemas and a validation step. Use this general pipeline when no specialized template fits.

Example inputs

  • A vendor purchase order PDF, extracting PO number, line items, and totals.
  • A signed contract, extracting parties, effective dates, and renewal terms.
  • A scanned expense receipt, extracting merchant, date, and amount.

Last updated: Jul 3, 2026