Sami Ullah

Owned end-to-end ML developer experience on Run:ai, supporting ~10,000 jobs/week across interactive experimentation, distributed training, and production model hosting on multi-GPU clusters. Developed runtime security compliance scanners and built CI/CD pipelines for compute resource rollout.

Optimized production computer vision inference pipelines, achieving 33% latency improvement for production detection models. Packaged and validated optimized models for deployment using NVIDIA Triton Inference Server.

Owned design and deployment of an agentic RAG system with SQL-backed grounding, achieving near-zero hallucination for agronomy decision support. Led development of a computer vision stand-count pipeline, reducing MAE from 15 to 2.4 and deploying models to production.

Optimized large-scale cloud data lake querying workflows, reducing query latency by 50% and data scanned by 70%. Built and maintained CI/CD pipelines for ML models and developed secure ML APIs with JWT-based authentication.

Developed and deployed production-grade ML models using PyTorch, improving model accuracy by 15% across core use cases. Expanded training datasets by 30% and improved labeling accuracy by 20% using generative models and advanced image processing pipelines.

Led development of a real-time semantic segmentation system for crop and weed detection in collaboration with industry partners. Optimized model architectures, reducing memory usage by 91 MB, parameters by 22M, and FLOPs by 116B with less than 1% accuracy loss.

Developed an anomaly detection system using audio and video signals in a semi-supervised setting to enhance surveillance system output. Implemented NLP-based text pattern analyzer to detect fraudulent claims of authorship and developed an end-to-end video analysis pipeline for brain tumor detection using CNN.

Activities and societies: Dean's Honor List Award, Two Teaching Assistantships, UR Graduate Scholarships

Master of Electronic Systems Engineering with a specialization in Deep Learning and Computer Vision. Awarded Dean's Honor List, teaching assistantships, and UR Graduate Scholarships.

Activities and societies: Three Gold Medalist for best performance, six times top in the batch of 80 students (Received Financial Award)

Bachelor of Science in Electrical Engineering. Received three Gold Medals for best performance and was six times top in a batch of 80 students.