Ahmad B.

I build production computer vision, generative AI, and multimodal ML systems — from avatar synthesis and real-time video to large-scale data pipelines. 4+ years of production experience shipping CV/AI at scale. Currently a Machine Learning Engineer at text-to-video generation company, working on avatar generation, video data pipelines, and AI proctoring. Previously a Research Assistant at the Intelligent Machines Lab, with a CVPR-submitted paper on 3D point cloud upsampling. Core expertise:  • Generative AI & video synthesis (Stable Diffusion, AnimateDiff, SVD, avatar pipelines) • Large-scale video data processing (Apache Spark, Argo Workflows, 10K+ hour pipelines) • Real-time CV systems — object detection, tracking, proctoring, matting (YOLOv11, SAM2, OpenCV) • 3D vision & NeRF, camera pose estimation, point cloud processing • Model deployment & optimization (ONNX, TensorRT, Docker, AWS) • Vision-language models & agentic AI (GPT-4o, Qwen-VL, LangGraph) — available for VLM prototyping Recent results:  • Redesigned company's core generation pipeline — improved results by 20%, driving a 30% lift in user engagement. • Architected a distributed video processing pipeline (Spark + Argo) that cleaned 10,000+ hours of raw video for generative model training. • Built a real-time AI proctoring system combining deepfake detection, gaze tracking, and facial behaviour analysis for interview integrity verification.
 Stack: Python, PyTorch, OpenCV, HuggingFace, LangGraph, Spark, Argo, ONNX, Docker, AWS.
 I work best with founders and teams who want a technical partner — I ask the right questions, flag risks early, and ship clean, documented code.
 Send me a message with your project idea and I'll reply within a few hours with feasibility notes, approach, and a realistic timeline.