Hi, I'm Behzad Hassan
AI & Computer Vision Engineer specializing in building end-to-end intelligent systems, real-time visual recognition pipelines, and high-performance production deployments.
About Me
AI/ML Engineer and Computer Vision Specialist specializing in Python, machine learning, and full-stack web development. My passion lies in bridging the gap between cutting-edge AI models and production-ready applications.
Recently concluded comprehensive research on benchmarking Vision Language Models (VLMs)on Pakistani medical imaging data. Now focused on engineering scalable, production-ready AI systems and full-stack solutions.
OS:
Location:
Role:
Status:
Email:
Technical Expertise
Scroll to explore my skills
AI / Machine Learning
Computer Vision
Full-Stack Development
Selected Projects
Model outputs from my workstation
LEVIR-CD Change Detection
Vision Language Model Benchmark for Change Detection. A human-in-the-loop evaluation platform for reviewing Qwen2-VL outputs on bi-temporal satellite imagery. Built for the LEVIR-CD dataset.
custom zones · object tracking
Real-time perimeter security system using YOLOv8, ByteTrack & FastAPI. Includes a Next.js dashboard for custom zone alerts.
accuracy: 78.4% · 5 models · 3 modalities
Evaluating 5 VLMs on Pakistani medical imaging modalities: Chest X-rays, CT scans, and Brain MRIs.
video streaming · telemetry
High-performance CCTV surveillance and people-tracking app. Features a FastAPI AI backend and Next.js dashboard for real-time video & telemetry.
LCP: 0.8s · Lighthouse 98
Premium e-commerce frontend — fluid swipe navigation, hover-zoom inspection, minimalist design system.
mIoU: 82.1 · 19 classes · real-time
Real-time semantic segmentation of urban scenes. SegFormer-B5 Transformer on Cityscapes.
accuracy: 94.8% · 96K records
AI Disease Prediction — Random Forest on 96K+ records. Personalized treatment plans.
track_id: stable · 30fps
Biometric fingerprinting for individual identification in video sequences. Real-time pipeline.
keypoints: 54 · 33 body + 21 hand
Reconstructs & tracks 3D human pose + hand landmarks in real-time. 54 total keypoints.
precision: 96.2% · cosine similarity
Facial recognition via FaceNet embeddings. Ranked similarity scoring against target directories.
landmarks: 468 · 3D mesh · 60fps
468-point 3D facial mesh in real-time. Precision tracking for AR and facial analysis.
Publications & Guides
Technical writing, field guides, and research papers
Get In Touch
Have a project in mind or just want to say hi? My terminal is always open.