Muhammad Saad

Data Science Intern (R&D) building an event-driven cutting-stock optimisation service on Microsoft Azure, integrating NVIDIA cuOpt to efficiently process industrial optimisation tasks at scale.

Muhammad is developing and optimising algorithms to solve complex cutting-stock problems — a core challenge in the metals industry where raw bars of steel must be cut to customer-specified lengths with minimal waste. His work involves designing a cloud-native pipeline on Azure: an HTTP API built with Azure Functions receives cutting jobs, queues them via Azure Service Bus, and a Container Apps Job picks them up and runs the solver. The system uses KEDA event-driven autoscaling so compute resources scale to zero when idle, keeping costs low.

A key part of the work is benchmarking: Muhammad implements classical heuristics such as First Fit Decreasing as a baseline, then compares their results against NVIDIA cuOpt — a GPU-accelerated optimisation engine — to quantify the performance and material savings achievable with AI-powered solving. On representative test cases, cuOpt achieves near-optimal solutions that reduce bar consumption by up to 17% compared to the heuristic baseline, translating directly into reduced material waste and cost for manufacturers.

Alongside solver development, Muhammad performs data analysis tasks including data cleaning, preprocessing, and validation to ensure high-quality inputs and reliable optimisation outputs. He follows engineering best practices throughout: the codebase is covered by an automated CI/CD pipeline with ruff linting and an 80% test coverage gate enforced on every pull request.

Muhammad is studying Computer Engineering and brings a strong interest in the intersection of cloud infrastructure, operations research, and applied AI. At GoSmarter he works within the R&D function, contributing to the technical foundation of products that help manufacturers make better use of their raw materials.