Document AI
Invoice Parser with LayoutLM
A document-intelligence system that parses and extracts structured data from unstructured PDF invoices across 12 vendor formats with minimal manual intervention.
- Industry
- Finance operations
- Status
- Delivered
- Outcome
- 94.7% extraction accuracy across supported vendor formats
Overview
A document pipeline that turns supplier PDF invoices into structured, validated records ready for accounting entry.
The challenge
Invoices arrived in twelve different layouts and were keyed by hand, which made accounts payable slow and error-prone.
Architecture
- AWS Textract for OCR and layout extraction
- LayoutLM models trained on labelled vendor formats
- FastAPI service exposing extraction and confidence scores
- Review queue for low-confidence fields before posting
Capabilities delivered
- Structured extraction across 12 vendor formats
- Field-level confidence scoring
- Exception review workflow
- Downstream posting into finance systems
Outcome
94.7% extraction accuracy across supported vendor formats.
More work