⭐ Featured Project

AI Invoice OCR & Billing Discrepancy Detection

Applied AI-based OCR and analysis to vendor invoices, identifying ~$9K in billing discrepancies and improving financial oversight.

Technologies & Tools

OpenAI OCRPythonPDF Processing

📊 Impact: ~$9K recovered from billing discrepancies

Problem

Tech Goes Home processes invoices from multiple vendors for devices and services, but invoice validation was largely manual and time-intensive, making it difficult to reliably detect billing errors at scale. Small discrepancies could easily go unnoticed, accumulating into significant financial loss. Invoice review relied on manual visual checks, line-item discrepancies were difficult to catch consistently, no scalable mechanism existed for cross-invoice validation, and financial leakage risk increased as invoice volume grew.

My Role

I owned the solution end-to-end, including designing the AI-based invoice analysis approach, implementing OCR and structured data extraction, normalizing invoice data for comparison, defining discrepancy detection logic, and validating findings and quantifying financial impact. This was a value-driven analytics initiative, not an experiment.

Solution

Designed an AI-powered invoice analysis workflow using OpenAI's OCR and language understanding capabilities to automatically extract, normalize, and analyze invoice data. This system surfaced 10 billing discrepancies totaling approximately $9,000, directly recovering funds and improving financial oversight.

Architecture

High-Level Data Flow

1

High-Level Data Flow: (1) Invoice documents ingested (PDFs/scans)

2

OpenAI OCR extracts text and structured fields

3

Line items and totals normalized

4

AI-assisted logic compares billed amounts against expected values

5

Discrepancies flagged for review

6

Findings validated and reported. Used OCR + language models to handle inconsistent invoice formats, avoiding rigid template-based parsing. Each discrepancy included traceable invoice references enabling finance teams to confidently act on findings.

Key Design Decisions

🔹AI for Unstructured Data - Used OCR + language models to handle inconsistent invoice formats, avoiding rigid parsing
🔹Explainability - Each discrepancy included traceable invoice references for confident action
🔹Targeted Scope - Focused on high-value invoices first to maximize ROI

Results

  • Identified 10 billing discrepancies totaling ~$9,000
  • Improved confidence in vendor billing accuracy
  • Reduced time spent on manual invoice checks
  • Demonstrated immediate ROI from applied AI
  • Established repeatable model for future financial audits

Technologies Used

OpenAI OCRPythonPDF ProcessingFinancial AnalysisPattern Recognition