Case study · AI / ML
A document AI pipeline that reads invoices and receipts, validates the data and posts it to accounting software for review.
- Client
- An outsourced accounting firm processing thousands of documents a month
- Industry
- Accounting services
- Region
- Europe
- Duration
- 10 weeks
- Team
- 2 AI engineers, 1 back-end engineer
Staff keyed invoice data by hand during month-end close, which meant overtime, late reports and typing errors.
80%
of documents processed with no manual edits
3 days
faster month-end close
−90%
data entry hours
Step 001
Document study
Sampled 2,000 documents across suppliers, languages and formats.
Step 002
Extraction
LLM extraction with schema-validated output, plus rules for totals, VAT and duplicates.
Step 003
Review queue
Low-confidence fields routed to a simple review screen instead of the whole document.
Step 004
Integration
Approved entries posted straight to the accounting system via its API.
- Python
- Google Gemini
- Pydantic
- FastAPI
- PostgreSQL
- Celery
- Docker
Tell us about it and we'll reply with how we'd approach it, usually within one business day.