Case Study

Blueprint Analyzer

AI-powered blueprint analysis that turns architectural drawings into demolition takeoff data in seconds.

PythonFastAPIOpenCVPydanticReactTypeScriptViteTailwind CSSAzure OpenAIGPT-5.2HTML5 Canvas

The Challenge

Environmental Demolition Group (EDG) was manually tracing vector points on architectural blueprints to calculate square footage, wall lengths, and material quantities for demolition scoping. It's slow, error-prone, and doesn't scale. The hard part: blueprints are messy — furniture overlaps walls, dashed lines sit on solid ones, hallways bleed into rooms through open doorways, and the drawing scale is buried in fine print.

The Approach

I built a dual-engine pipeline that pairs computer vision with AI reasoning — OpenCV handles geometry, Azure OpenAI's GPT-5.2 handles understanding. The backend ingests a blueprint image, runs it through a multi-stage preprocessing pipeline (green line removal, adaptive thresholding, intensity filtering to separate walls from furniture), then detects room contours and calculates pixel areas. Simultaneously, the LLM reads the drawing scale text and infers room types from fixtures and layout context.

The frontend renders the blueprint on a canvas with colored room overlays, a live data table, and a full demolition takeoff summary — all recalculating instantly when the user adjusts the scale slider, drags a room boundary, or adds/deletes rooms. The key design decision was keeping humans in the loop: AI accelerates the workflow, but the user always has final say.

Blueprint upload interface showing file processing settings with scale detection and room detection options

Upload interface — blueprint ingestion with configurable CV and AI detection settings

Blueprint canvas with colored room overlays and numbered room detection, sidebar showing detected rooms table with areas and perimeters

Room detection — 30 rooms identified with colored overlays, live area and perimeter calculations, draggable boundaries

Demolition takeoff report showing room schedule, line-item cost estimates by category, and grand total

Demolition takeoff — room schedule with line-item cost estimates across flooring, walls, doors, ceiling, and totals

Key Results

43 rooms detected, 339 wall segments identified in under 0.1s
CV processing only — geometry extraction is near-instant on standard architectural drawings.
18 rooms auto-labeled by type from fixtures alone
Bedroom, bathroom, kitchen, etc. — inferred purely from fixture layout, no text on the blueprint required.
Scale extraction + room labeling in ~17s end-to-end
Replacing what took manual tracers significantly longer per blueprint.
Instant client-side recalculation
Adjusting scale or boundaries updates all square footage, linear feet, and takeoff costs with zero backend round-trips.

← Back to Portfolio