Overview

A large financial services back-office team was receiving 400–600 documents per day via email, fax-to-PDF, and file drop. Documents included account statements, wire instructions, compliance filings, and client correspondence — each requiring classification, extraction, and routing to different downstream teams.

What We Built

• Multi-source ingest: IMAP polling, SFTP watch folder, S3 upload webhook
• Rule-based and ML-assisted document classification (12 document types)
• Structured data extraction for key fields (account numbers, amounts, dates)
• Routing engine: auto-assign to team queues based on type + extracted metadata
• Exception queue for low-confidence documents requiring human review
• Audit trail: every document's processing history, actor touch, and final disposition

Stack

PHP 8.2, Python (extraction service), PostgreSQL, RabbitMQ, AWS Textract, S3.

Outcome

Automated handling of 78% of inbound documents. Manual review time reduced by 60%. SLA compliance improved from 71% to 97%.