Unstructured Data Extraction
Adeptia's Unstructured Data Extraction capability uses AI to read documents in their original, unstructured form and convert their contents into clean, structured data, making it straightforward to incorporate document-based information into your integration workflows.
What you can do
With this capability, you can upload a document, let the AI identify and extract the relevant information it contains, and receive that information back in a structured format that downstream systems can consume directly. This removes the need to manually parse or reformat document content before it can be used in an integration.
The capability supports two complementary operations:
| Operation | What it does |
|---|---|
| Field discovery | Analyzes a document and returns a list of the data fields it contains, so you can see what information is available before deciding what to extract |
| Structured extraction | Reads a document and produces a structured output containing the values for the fields you care about |
Supported document types
You can submit documents in a range of formats, including PDF, common image formats (PNG, JPG, JPEG, TIFF, BMP), Word documents, Excel spreadsheets, PowerPoint presentations, HTML files, and plain text files.
What you configure
When using this capability, you provide:
- The document to process: the file you want the AI to read and extract data from.
- The document reading approach: you choose how the document is parsed. For PDFs, you can select from available readers including a general-purpose reader and Azure AI Document Intelligence. For image-based documents, you can choose a general-purpose image reader or Azure AI Document Intelligence, which also supports specialized pre-built models for document types such as invoices and receipts.
- The pages to process: for multi-page PDFs, you can process all pages or specify a subset of pages by number or range.
- The fields to extract (optional): if you already know which fields you want, you can supply their names and the AI will extract only those fields and their values. If you do not supply field names, the AI extracts all relevant information it finds.
- The output format: you choose whether the structured output should be delivered as JSON or CSV.
- Additional instructions (optional): you can supply supplementary instructions to guide how the AI formats or organizes the extracted output.
How it fits into your integrations
Unstructured Data Extraction is designed to sit at the front of an integration flow where the source data arrives as a document rather than a structured feed. Once the AI has extracted and structured the document content, the resulting JSON or CSV output can flow directly into the rest of your integration, just like any other structured data source.
For multi-page documents, the AI processes each relevant page and collects the results page by page.
Document reading options
The platform offers several ways to read a document, and the right choice depends on the nature of your files:
| Reading approach | Available for |
|---|---|
| PyPDFium2 | PDF files |
| Unstructured | PDF files and image-based documents |
| Azure AI Document Intelligence (pre-built document model) | PDF files and image-based documents |
What to expect from the output
When you request JSON output without specifying a custom structure, the AI organizes all extracted data into a well-formed JSON object. For CSV output, the AI delivers the extracted data in CSV format.
If the document content triggers a content safety filter, or if the credentials used to connect to the underlying AI service are invalid, the platform returns a clear error message so you can take corrective action.