Soroush F.
Machine Learning Engineer @Hologen
Signup · Get unlimited contacts
WORK HISTORY
Machine Learning Engineer @Hologen
London, GB
Led large-scale optimisation of distributed ML training systems across multi-node GPU clusters, delivering up to 30%+ throughput gains and significantly reducing performance variance in long-running jobs.• Architected and rolled out production-grade cluster scheduling and resource management policies (SLURM-based) with zero downtime, improving utilisation, fairness, and cost efficiency across teams.• Designed and implemented advanced data pipeline and caching strategies that reduced GPU idle time, cut throughput dips by over 90%, and improved training stability at scale.• Directed multi-environment benchmarking initiatives (25+ experiments) across heterogeneous GPU hardware, enabling data-driven infrastructure decisions and scaling from single-node to 20+ nodes.• Built secure, fully offline ML deployment pipelines using self-contained containerised environments, enabling air-gapped inference with 100% validation success in restricted infrastructure settings.• Established performance observability standards using deep profiling and telemetry (e.g, NVIDIA Nsight), transforming opaque distributed systems into measurable, optimisable infrastructure.
EDUCATION
University of Surrey
Doctor of Philosophy - PhD, Electrical and Electronics Engineering
Shiraz University
Bachelor of Science - BS, Computer Hardware Engineering
University of Tehran
Master's degree, Information Technology
ABOUT SOROUSH F.
I am an AI specialist with a Ph.D. in Vision, Speech, and Signal Processing from the University of Surrey and over eight years of experience in machine learning, deep learning, and computer vision. At Terminal Industries, I optimise and deploy vision-based AI models, leveraging Generative AI (GenAI), foundation models, multi-modal learning, and LLMs to enhance model adaptability and efficiency. My expertise spans model optimization, quantization, and low-latency inference using NVIDIA Triton, ONNX, TensorRT, and AWS (EC2, S3, Lambda) while integrating asynchronous pipeline execution, ClearML observability, and auto-scaling ML workflows for high-performance AI systems.Previously at Samsung R&D, I led the development of efficient computer vision models, including image captioning, scene classification, and real-time object detection, achieving 10x latency reduction and 8x memory savings with hardware-aware pruning, distillation, and Transformer-based optimizations for edge AI. My experience extends to MLOps, CI/CD, and scalable AI deployment, ensuring models transition seamlessly from research to production.I am proficient in Python, PyTorch, TensorFlow, C++, SQL, and cloud-native AI architectures, with expertise in NVIDIA Nsight, Compute Profilers, and distributed AI training. With 10+ peer-reviewed publications, my research spans anomaly detection, self-supervised learning, robust AI deployment, and domain adaptation. Passionate about scalable and explainable AI, I drive innovation in next-generation AI systems that power real-world applications.Key Skills & Expertise:* AI & ML: Deep learning (CNNs, transformers), anomaly detection, domain adaptation, model optimization.* Deep Learning Frameworks: PyTorch, TensorFlow, ONNX, NVIDIA Triton.* Model Optimization: Quantization, pruning, export to ONNX, reducing latency/memory usage.* Cloud & Edge Deployment: AWS (EC2, S3), scalable model pipelines, edge device optimization.* Programming Languages: Python, C/C++, Java, R, SQL/MySQL, PHP.* Data Science & Tools: Pandas, NumPy, SciPy, Matplotlib, scikit-learn, IBM SPSS Modeller.* Performance Profiling: NVIDIA Nsight and Compute Profilers.* DevOps & CI/CD: Git, ClearML, JIRA, CI/CD pipelines for automated ML workflows.* Research & Publications: 10+ peer-reviewed papers in leading journals on anomaly detection, robust AI models, and optimization techniques
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.