2026prototype

Lingua Franca

A local pipeline that turns inspection emails, spreadsheets, images, and PDFs into validated routing decisions.

Built in 24 hours at GridStorm Hacks 2026, Lingua Franca accepts inspection screenshots, spreadsheets, images, and PDF reports, then converts them into structured records.

I built the ingestion and extraction layer with OpenCV, Tesseract, openpyxl, and pypdf. A local Mistral model extracts fields; Pydantic validation, completeness scoring, and deterministic rules decide where each record goes and when a person should review it.

Lingua Franca teammates developing the document extraction pipeline during GridStorm Hacks

Lingua Franca teammates working together with laptops during the hackathon

Lingua Franca evaluation chart comparing extraction and routing accuracy across input file types