Industrial AI / Computer Vision · Hackathon MVP — 1st Place 🥇
RootVision AI
An AI-powered visual inspection platform for metal garage doors that detects defects, maps them to the responsible production stage, and generates explainable root-cause recommendations.
Overview
RootVision AI is an AI-powered visual quality inspection platform for metal garage doors, built solo in 72 hours for the AUTOTHON hackathon at King Faisal University in collaboration with Al-Barrak Garage Doors Factory. Instead of only saying 'defect detected,' it answers four explainable questions for every door: What is the defect? Where is it located? How severe is it? Which production stage is responsible — and what corrective action is recommended?
The Problem
Traditional visual inspection in metal manufacturing relies entirely on human judgment — slow, inconsistent, and unable to trace a defect back to the production stage that caused it. Inspectors can identify 'there is a dent' but cannot determine whether it originated during forming, handling, or welding, making root-cause analysis and systematic process improvement nearly impossible.
System Flow & Architecture
Explainable AI Flow
RootVision Score Formula
The score is a weighted sum of four independent sub-scores (0–100 each):
1. Visual Defect Score × 0.40 — defect count, size, location, and type severity.
2. Specification Compliance Score × 0.25 — validates against the Al-Barrak standard (1 m / 90° / ≥500 Lux).
3. Motion Signature Score × 0.20 — door open/close timing and smoothness vs. the golden reference curve.
4. Root Cause Confidence × 0.15 — how clearly the rule engine traces the defect to a specific production stage.
Status bands: ≥ 85 → PASS · 60–84 → REWORK REQUIRED · < 60 → REJECT.
Demo door G-2026-014: 70·0.40 + 60·0.25 + 88·0.20 + 85·0.15 = 73 → REWORK REQUIRED.
Business Value
Before vs. After
Manual inspection relies on a human eye taking 5–10 seconds per area, producing only a pass/fail verdict, applying the standard from memory, and leaving no data trail. RootVision AI processes each image in ~0.42 seconds and outputs defect type, location, severity, likely production stage, and recommended corrective action — with the Al-Barrak standard encoded and every result stored as a KPI.
Who Benefits
Quality inspectors receive faster, objective verdicts with machine-readable reasons instead of gut feel.
Production supervisors know exactly which stage to investigate and fix, not just that something went wrong.
Plant managers gain measurable SAR savings and quality KPIs — pass rate, rework rate, rejection rate — to track process improvement over time.
ROI & Cost Savings
Key Assumptions
120 doors inspected per day (one production line) · SAR 1,500 door value · SAR 2,000 cost per escaped defect (return, site visit, rework, reputation) · Baseline escape rate 4% → improved to 1% with 90%+ automated detection.
Daily Savings Breakdown
3.6 fewer escaped defects per day → SAR 7,200 saved daily.
15 dents caught before painting instead of after (SAR 450 saving per door) → SAR 6,750 daily.
Total daily savings: SAR 13,950 per line.
Annual Return (per line)
Annual gross savings: ~SAR 4,185,000.
Annual platform cost (cameras + compute + license + maintenance): SAR 300,000.
Net annual benefit: SAR 1,860,000.
ROI: ~620% · Payback period: ~2 months.
Market Opportunity
Why Now
Commodity cameras are cheap enough for any production line. YOLOv8 reaches 90%+ accuracy on a focused defect class with a small dataset. Explainable output — location, severity, likely stage, recommended action — makes results trustworthy enough to act on without a data-science team.
Competitive Edge
Manual inspection is slow and subjective; RootVision AI is consistent and instant. Generic defect detectors output only 'defect detected'; RootVision AI explains what, where, severity, stage, and action. Heavy machine-vision rigs are expensive and fixed-purpose; RootVision AI runs on commodity cameras and a laptop.
Go-to-Market
Pilot on one garage-door line at Al-Barrak — prove the SAR savings in the ROI model. Add scratch, paint, and missing-screw classes. Expand to adjacent sheet-metal, gates, and appliance manufacturing. Aligned with Saudi Vision 2030 industrial-localization goals; SAR-denominated ROI resonates directly with regional manufacturers.
Key Features
- YOLOv8 dent detector — Precision 92% / Recall 91% / F1 91.5% at 0.42s/image
- Explainable AI Rule Engine: defect + location → likely production stage + recommended action
- RootVision Score (0–100): Visual × 0.40 + Compliance × 0.25 + Motion × 0.20 + RCA × 0.15
- Al-Barrak visual standard compliance validation (1 m / 90° / ≥500 Lux)
- Motion signature analysis: open/close curve vs. golden reference
- 4-screen monochrome dashboard (Home, Defect Detail, Motion Signature, Reporting)
Outcomes & Impact
- Trained YOLOv8 dent detector achieving Precision 92% / Recall 91% / F1 91.5% at 0.42s per image.
- Won 1st place (7,000 SAR prize) at AUTOTHON hackathon — King Faisal University × Al-Barrak Factory.
- Designed and built the complete 3-layer system (React frontend + FastAPI backend + AI layer) solo in 72 hours.