All Projects
AI · Document Automation · Internal Tooling

Inspection Digitizer

Replaced manual, hours-per-report digitization of complex industrial inspection PDFs into structured Excel deliverables with an AI-driven pipeline, cutting turnaround from hours to minutes per report.

Overview

An internal tool built for an industrial inspection company to automatically extract and digitize complex, multi-page PDF inspection reports into the company's standard Excel deliverable format — replacing manual data entry with an AI-driven document pipeline that still gets independently checked before anything is trusted.

Key Features

  • Single-report and large-batch processing modes, including bulk AI processing at reduced cost for hundreds of reports in one run
  • A second, independent AI pass that cross-verifies every extraction against domain rules before it is accepted
  • Checkpointed extraction pipeline that resumes cleanly after an interruption without re-running expensive AI processing
  • Automatic sanitization of AI-extracted data to guarantee clean, spreadsheet-compatible output every time
  • Local encrypted credential storage so the tool never requires exposing API keys in plaintext

Architecture

CLI-driven Python pipeline: AI-based PDF extraction, checkpointed intermediate storage, an independent AI cross-verification pass, and automated Excel deliverable generation

What I Learned

Working with LLMs on messy real-world PDFs taught me that the model call is the easy part — most of the engineering is defensive: catching truncated output, stripping invisible characters that silently corrupt a spreadsheet, and building a verification pass that doesn't just trust the first answer it gets.