Stud Finder AI

About this project

Stud Finder AI is an agentic computer vision and multi-agent AI system designed to analyze architectural PDF blueprints and compile deterministic lumber material takeoffs with high precision.

The Challenge

Performing lumber and framing material takeoffs from architectural PDF blueprints manually is a slow, error-prone task that can lead to either costly project delays or expensive material surpluses. Estimators spend hours tracing and counting studs, headers, and plates across dense drawings.

The Solution

We engineered an automated pipeline utilizing advanced computer vision models and structured AI agents. The system reads architectural blueprint layers, identifies structural walls and framing elements via OpenCV, and applies custom rule-based parsing via Pydantic-AI and Gemini 2.5 Pro to output exact material counts and pricing estimates.

Automated PDF blueprint parsing and

Automated PDF blueprint parsing and structural wall extraction.

Computer vision scanning for detecting

Computer vision scanning for detecting door, window, and wall dimensions.

Multi-agent validation pipeline reconciling visual

Multi-agent validation pipeline reconciling visual detections with building code requirements.

Real-time server-sent events (SSE) indicating

Real-time server-sent events (SSE) indicating extraction progress and model confidence.

AI Intelligence

Multi-agent system powered by Gemini 2.5 Pro and Pydantic-AI that coordinates image parsing, lumber conversion calculations, and structural validation checks.

System Architecture

Python backend utilizing FastAPI and OpenCV for asynchronous vision processing, delivering real-time logs via SSE streams to a modern React dashboard.

Database & Caching

PostgreSQL handles user files and final takeoffs, while Redis caches temporary extraction results and active queue states.

Messaging

Server-Sent Events (SSE) provide live progress bars, while Twilio SendGrid delivers detailed Excel spreadsheets once the takeoff is finalized.

Frontend

Backend

Deployment

Backend deployed in Docker containers on Google Cloud Run with GPU-accelerated processing instances.