Arm
Cambridge
I completed a six month industrial placement at Arm as a GPU Hardware Engineer in the Verification Methodology team. My work focused on internal engineering tools and AI assisted verification workflows. Early in the placement, I contributed to an internal AI Knowledge Retrieval Tool (KRT), improving its evaluation framework, extending document ingestion support, and helping prepare the application for wider deployment. Later, my work moved closer to verification methodology, where I contributed to AI assisted review and signoff tooling for GPU verification. This involved processing signoff evidence, generating structured review packs, and developing agentic workflows to support unit level testbench correctness analysis. Across both projects, I gained practical experience of software development, GPU verification workflows, deployment, authentication, and building maintainable engineering tools.
My role was GPU Hardware Engineer within Arm’s Verification Methodology team. I contributed to two main engineering projects. For KRT, I improved the evaluation framework, added support for additional document formats, and helped implement changes required for wider company rollout. For the GPU verification signoff project, I worked on tooling that processes unit level signoff evidence, generates deterministic review packs, supports evidence parsing, and applies agentic code review to verification questions. I also independently proposed and developed IP-ASA, an AI assisted code review prototype for time consuming signoff analysis. I defined its scope, inputs, workflow and output structure, using a five stage agent process: inventory, analysis, skeptic, judge and synthesizer. My work helped make internal tools more reliable, accessible, structured and useful to engineers.
During the placement I used and developed skills across software engineering, AI tooling and hardware verification. I worked with Python, JSON and YAML data structures, CLI workflows, automated tests, repository restructuring and shared codebases. I also gained experience with Elasticsearch, Kibana, internal deployment environments, OIDC authentication and access control. Through my verification work, I learned more about GPU architecture, UVM testbenches, unit level signoff requirements and the types of evidence needed to assess testbench correctness. I also developed experience applying agentic AI models to code and verification analysis. A key lesson was that engineering solutions must be maintainable, reviewable and well documented, not simply functional, especially when tools are used by other engineers or revisited months later.
The placement strengthened my communication, collaboration, independence and task management skills. Working in industry showed me the importance of clear communication through stand-ups, one-to-one meetings, merge request descriptions, documentation and asking for help when needed. I became more confident making decisions and managing open ended tasks independently. This was especially important during the signoff project, where requirements and project direction changed quickly and I had to revise assumptions and adapt my work. I also developed stronger initiative by identifying a time consuming verification problem and turning the idea into the IP-ASA prototype. Overall, I learned how to contribute effectively within a professional engineering team while balancing technical work, changing requirements, collaboration and clear documentation.