Engineers spend hours manually reading through technical drawings — P&IDs, single line diagrams, equipment layouts — to extract asset tags one by one. Rosary Vision automates that entirely. Upload a PDF, click extract, and download a clean CSV of every equipment tag, instrument label, and pipe spec on the page.

As the founding developer at Rosary Labs, I built this platform end-to-end. The extraction engine combines three AI techniques — object detection, OCR, and vision language models — processing each drawing as an image grid with multi-rotation consensus voting to handle text in any orientation. The web application is built on Django with Google and Microsoft OAuth, real-time bounding box editing using Fabric.js, an Excel-like Bill of Quantities verification workflow, background job processing, and a Docker-based deployment pipeline on Azure.

I delivered 25 features spanning the full stack — from the initial extraction scripts and ground-truth annotations to the production-ready web app with admin dashboards, feedback systems, and a downstream integration API. The project has since been handed off to the team who continue to evolve the product.