FormDesk

Challenge Supplier onboarding and document updates require accurate structured records. Reviewers need to check extracted values against the source before those values enter a spreadsheet or operational system.

My role AI Engineer at Reallytics, July–December 2025. Developed as a two-person team; this case study describes my full-stack application contribution without implying sole authorship.

Solution I built a document-processing application that reads text and scanned PDFs, presents extracted fields beside highlighted source evidence, and lets reviewers confirm or correct each value. Email and date validation, batch processing and a review-status trail support the review workflow. Confirmed data can be exported to CSV or Excel.

Documented outcome Delivered an intake → extraction → evidence review → export workflow. The portfolio screenshots document intake, field review and batch-processing interfaces. The practical result is a reviewable path from source documents to structured exports; processing-time savings and extraction-accuracy improvements have not yet been measured.

Technology and context Python, JavaScript, HTML/CSS, pdfplumber and local OCR. The public portfolio edition uses fictional documents and a redesigned brand to protect client confidentiality. It demonstrates the application rather than disclosing original client data.

Reads text and scanned PDFs for supplier onboarding, contact updates, and service requests.

Displays extracted values beside highlighted source evidence.

Lets reviewers confirm or correct fields before exporting to CSV or Excel.

Includes batch processing, email and date validation, and a review-status trail.

Built with Python, JavaScript, HTML/CSS, pdfplumber, and local OCR.

This portfolio edition uses fictional documents and a redesigned brand to protect client confidentiality.

Built as a two-person team.