AI-driven RPA MVP for African SME back-office automation — built from the thesis 'Digitizing Back Office Operations Through AI-Driven Robotic Process Automation'
# Zola Pay Automation Console
An MVP implementation of the AI-driven Robotic Process Automation (RPA) framework proposed in the thesis **"Digitizing Back Office Operations Through AI-Driven Robotic Process Automation"** (Kellen Karwitha Kinyua, MSc IT, Strathmore University).
The thesis proposes a lightweight, AI-integrated RPA framework for African SMEs in **digital financial services and logistics**, targeting back-office processes such as invoice/payment reconciliation and customer/document verification, measured against turnaround time, automation rate, accuracy, and cost reduction (Table 3.1).
This repository is a working demonstration of that framework: a real OCR → AI classification → field extraction → business-rules pipeline processing scanned documents for **Zola Pay**, a fictional pan-African merchant payments platform modelled on the business shape of firms like Cellulant or Flutterwave (merchant collections & payouts across multiple African markets). Every number on the dashboard is computed live from the database — nothing is hard-coded.
> See `docs/THESIS_MAPPING.md` for how each thesis objective and metric maps to a concrete feature, and `docs/DEMO_SCRIPT.md` for a walkthrough script for a defense panel or a business-owner demo.
## What it does
Two back-office workflows, sharing one document-intake pipeline:
1. **Invoice & payment reconciliation** — a merchant invoice (PDF/image) is OCR'd, classified, has its fields extracted, then matched against a settlement ledger. Exact matches post automatically; mismatches or missing settlements route to a human review queue.
2. **KYC / onboarding verification** — a national ID or business permit is OCR'd and checked against completeness/expiry rules, with a risk level and approve/review decision.
A third document type (logistics waybills) is also classified and extracted, demonstrating the framework generalizes beyond the two primary workflows.
Every decision — automatic or human-resolved — is written …