docstrange_skill

Overview

This skill provides a wrapper around DocStrange by Nanonets for document extraction. It converts PDFs and images to markdown, JSON, or CSV and returns per-field confidence scores. Use it to OCR scanned documents, extract invoice or receipt fields, parse tables, or produce structured JSON from unstructured documents.

How this skill works

The skill sends files (or file URLs/base64) to the DocStrange extraction API and requests the desired output format (markdown, json, csv). It supports sync extraction for small documents and async queuing for large documents, returning a record_id to poll for results. You can include JSON field lists or schemas, request metadata like confidence scores or bounding boxes, and combine multiple formats in one call.

When to use it

• Automating invoice and receipt processing in accounting workflows
• Converting scanned contracts, bank statements, or forms into searchable markdown or structured JSON
• Extracting tables from PDFs into CSV for analytics
• Processing multi-page documents asynchronously to avoid timeouts
• Quick OCR of images where layout or coordinates (bounding boxes) matter

Best practices

• Use sync endpoint for documents <=5 pages; switch to async for larger documents
• Pass a JSON field list or schema for more accurate, typed JSON extraction
• Add include_metadata=confidence_score and review fields with score <80 manually
• Supply custom_instructions or prompt_mode to bias extraction toward critical sections
• Keep API keys in environment variables or a secure secrets store and rotate regularly

Example use cases

• Extract invoice_number, date, vendor, and total_amount into a typed JSON payload for ERP ingestion
• Convert a scanned contract into markdown and capture section hierarchy and bounding boxes for review
• Parse a retail receipt into merchant, date, total, and line items for expense reporting
• Convert PDF tables to CSV for bulk import into spreadsheets or BI tools
• Queue large back-office document batches with async extraction and poll for completion