← Back to Projects Address Extraction Review Tool for Scanned Documents
Extract address fields from scanned documents and route them through a review UI before export.
Categories
CVNLP
Tech Used
PythonOCROpenAIVision APIFlaskAPIHTMLCSSDockerAWS
Problem
Address blocks can be fragmented, noisy, or inconsistently formatted in real documents, creating errors in billing, shipping, and back-office workflows.
Approach
- Located candidate address regions using OCR and layout-aware heuristics
- Structured components with NER- and rule-based parsing logic
- Added a correction interface so users can review and fix uncertain outputs before export
Results
- Practical extraction flow for messy address data in scanned documents
- Human-in-the-loop review path for safer operational use
Demo Videos