OmniInspect — AI-Powered Visual Quality Control

OmniInspect is a multimodal AI system that solves one of manufacturing's most persistent problems: visual quality control at scale. Traditional approaches either rely on human inspectors who fatigue and miss defects, or rule-based computer vision systems that take months to configure and break with every product change. OmniInspect uses a 3-agent pipeline powered by Qwen2-VL-7B running on AMD Instinct MI300X: 1. Detection Agent — Analyzes enhanced images for any surface anomaly including cracks, scratches, contamination, holes, and deformation 2. Classification Agent — Classifies defect type, assigns severity (1-5), determines risk level (LOW to CRITICAL), and identifies probable root cause 3. Report Agent — Generates structured JSON reports with actionable recommendations: PASS, QUARANTINE, REJECT, or ESCALATE The AMD MI300X's 205.8GB of HBM3 unified memory is central to the pipeline — it allows full high-resolution images to be processed without downscaling, multiple vision models to be loaded simultaneously, and parallel batch inference across product streams. This is not possible on standard 24GB GPUs. Evaluated on the MVTec Anomaly Detection dataset (5,354 images, 15 industrial categories), OmniInspect achieves 80% accuracy zero-shot — improving from a 65% baseline through contrast and sharpness image enhancement preprocessing. The entire pipeline requires zero training data, zero labelling, and can generalize to any visual inspection domain out of the box. The system ships with a live industrial-grade web dashboard built on FastAPI, where users can drag and drop images and receive complete inspection reports in under 20 seconds.