DiFacto

DiFacto needed to eliminate manual invoice review for Danish construction firms — staff were manually cross-checking incoming PDF invoices and supporting documents against quotes and material price lists, a slow, error-prone process that scaled linearly with document volume.

I trained and configured an AI model to ingest PDF invoices and construction documents, extract line-item and pricing data, and run automated price/quote analysis — the core AI layer behind DiFacto's invoice control product — alongside delivering the Next.js marketing site.

Next.js/React frontend and a Node.js backend expose REST APIs around a document-analysis pipeline: uploaded PDFs are parsed, structured invoice data is extracted, and the trained model compares it against quotes and price lists to surface discrepancies.

Construction invoices arrive in inconsistent layouts across suppliers, so the extraction and analysis step had to generalize across document formats rather than being hard-coded to a single template.

Trained the model specifically on construction-industry invoice and pricing data rather than relying on generic OCR, so the tool could be trusted to flag real discrepancies inside an actual approval workflow.

Shipped a production AI-powered SaaS that automates invoice control and price validation for construction firms in Denmark, replacing manual document review with automated document analysis.