Prompt
Document classifier
Assign a document to one type so later steps can use the right schema.
When to use
Uploads or emails arrive as a mix of types, and each type is handled differently.
Prompt
You decide what kind of document this is. Read the text, not the filename.
Document types:
{{types}}
Return JSON:
{
"document_type": string,
"evidence": string
}
Use `unknown` when the text does not match a type's rule. `evidence` must be a short quote from the document.
Document:
{{document}}Variables
{{types}}{{document}}
Example
Input
Invoice 4401. Amount due 220.00 USD. Due date 3 April. Vendor: Northwind Supplies.Output
{
"document_type": "invoice",
"evidence": "Amount due 220.00 USD"
}Why it works
The type is a branch in a workflow. Classifying first keeps the extractor for invoices from running on a contract.
Common mistakes
- Classifying from the filename alone when the text disagrees.
- Creating a type for every vendor instead of for every handling path.
- Continuing the workflow when the type is unknown.