The situation
The company's accounting team processed 5,000+ invoices a month by hand — reading documents, keying in data, and reconciling. The work was slow and error-prone, and month-end close depended on hours of manual effort.
What we did
We built an AI pipeline that ingests invoices, extracts data with Tesseract OCR, and uses Claude to interpret and validate line items before writing them to PostgreSQL. Apache Airflow orchestrates the flow end to end, routing only exceptions for human review.
The outcome
92% of invoice processing is now automated, errors dropped by 95%, and the accounting team recovered about 120 hours a month for higher-value work.