WhatsApp AI Transaction Verification Agent for Payment Operations

The client processed payments across 50+ WhatsApp groups entirely through manual staff review. Each payment screenshot had to be individually opened, verified against records, and checked for duplicates or fraud. The process was slow, inconsistent at scale, and provided no audit trail. Fraudulent and duplicate submissions were passing through undetected, creating direct financial exposure. The operation had no automated mechanism to flag suspicious transactions before they were processed.

Built a three-container Docker Compose system deployed on DigitalOcean. The first container runs a Node.js Baileys WhatsApp bot that monitors all group chats and captures incoming payment screenshots. The second container runs a FastAPI OCR service using Tesseract as the primary extractor and Gemini Vision as the production fallback for low-confidence reads. The third container runs an n8n orchestration workflow that routes extracted transaction data through a Stripe API verification call and classifies each result as Valid, Fake, or Duplicate. Flagged transactions trigger real-time Telegram alerts. All results are logged to Google Sheets for audit. Security hardening includes read-only container filesystems, no-new-privileges flags on all containers, message-ID idempotency protection, bounded base64 input validation, duplicate suppression via a persistent store, and low-confidence OCR gating before any Stripe API call is made.

Eliminated manual payment screenshot review across 50+ WhatsApp groups entirely. The system operates 24/7 without human intervention, classifying transactions in real time. Fraud and duplicate detection now happens at the point of submission rather than after processing. A complete audit trail is maintained automatically in Google Sheets. Telegram alerts give the operations team instant visibility on any flagged transaction without requiring them to monitor groups directly.