OCR Fixture Code Detector

A web app for extracting lighting-fixture codes from electrical drawings.

Upload a PDF of an electrical lighting plan, and the app renders the page, runs OCR, and surfaces only the fixture codes you care about — LF1–LF15 plus LF7-X — each one tagged with a pixel bounding box drawn over the rendered drawing. Click a code to highlight its box, adjust or relabel boxes in-session if the OCR got something wrong, then export the results to JSON or CSV.

Live demo: https://ocr-fixture-code-detector.vercel.app

How it works

1. The PDF is rasterized to a high-resolution image on the server.
2. An OCR provider returns every recognized token with pixel-space bounding boxes.
3. A regex filter (^LF(\d{1,2}|7-X)$) keeps only the lighting-fixture codes and drops circuit numbers, EM / VS / D tags, and other noise.
4. The frontend overlays the boxes on the rendered drawing and lists the codes with counts and confidence scores.

Tech stack

• Backend: Python + FastAPI on Render, with PyMuPDF for PDF rendering.
• Frontend: React + Vite + TypeScript on Vercel.
• OCR: Google Document AI