Invoice OCR Extractor — Aniket Thakur

The Problem

Manual data entry from PDF and scanned invoices is labor-intensive, error-prone, and slow across accounting and procurement workflows.

Engineering Approach

Built a processing lifecycle that parses PDF/image uploads, extracts line items and header totals via OCR, and provides a verification dashboard.

What I Built & My Contribution

Created document upload validator, OCR parsing workers, fallback regex field extractors, and review dashboard.

Architecture & Data Pipeline

Document Upload → MIME/Size Validation → PaddleOCR Processing → Key-Value Extraction → SQLite Storage → Review UI

System Architecture

[ Invoice PDF / Image ] ──► [ MIME & Size Validator ]
                                    │
                                    ▼
┌────────────────────────────────────────────────────────┐
│     PaddleOCR Extraction & Regex Fallback Engine       │
│  - Vendor Name, Invoice #, Date, Line Items, Total Tax │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
           [ SQLite Database & Review Screen ]

Verified Outcomes & Metrics

Strict upload boundary enforcement; automated table extraction; visual human-in-the-loop review interface.

Verified Implementation

The upload service strictly constrains document extensions, MIME types, and file size, and exposes a transparent processing lifecycle. Verification here represents code audit and local inspection, distinguishing functional code paths from external cloud provider accounts.

Boundaries & Known Limitations

OCR quality varies with document resolution and non-standard layouts. No formal production accuracy SLA is claimed.