Rabia W.

I'm an AI/ML engineer specializing in computer vision. I build the systems that read documents people currently read by hand: engineering drawings, technical schematics, and security footage. Most recently I led a program extracting 32 classes of instrumentation and equipment symbols from Mechanical Drawings at 95.6% mAP across 10,000 drawings, replacing thousands of hours of manual takeoff. WHAT I DO • Document & drawing AI: P&ID and schematic symbol detection, OCR, field and table extraction • Object detection & tracking: YOLOv8, real-time inference, crowded and low-light conditions • Annotation programs: schema design, guideline authoring, team leadership, QA and agreement measurement • End-to-end delivery: data cleaning → annotation → training → evaluation → deployment SELECTED RESULTS • 32-class P&ID symbol extraction — 95.6% [email protected] across 10,000 drawings, 41ms per sheet • Unattended baggage & restricted-zone intrusion detection — 94.1% mAP, 20,000 frames, deployed on-premise and air-gapped • Facial attribute corpus for generative model localization — 10,000 images, 24 attributes, inter-annotator agreement raised from 0.74 to 0.91 HOW I WORK I've led annotation teams of five and owned the full pipeline on three programs. The hard part in vision projects is almost never the model, it's the label schema and the error analysis. That's where I spend my time, and it's why these numbers hold on held-out data rather than only in training. I work with confidential data regularly: on-premise deployments, air-gapped pipelines, and NDA-safe delivery practice. TOOLS Python · PyTorch · YOLOv8 · YOLOv11 · OpenCV · CVAT · Label Studio · Docker