Worked on expanding and refining the OpenVINO model architecture transformation verification framework within the huggingface/optimum-intel repository. Focused on increasing test coverage for architectures such as lfm2, lfm2_moe, qwen3_moe, and llama4, the work included introducing version guards to maintain compatibility with OpenVINO 2025.3.0 and later. Refactored the test configuration from TXT files to Python dictionaries, improving maintainability and reliability. Enhanced CI workflows to ensure comprehensive testing across multiple OpenVINO versions, reducing regression risk. Leveraged skills in Python, machine learning, and testing to deliver a more robust and scalable system for integrating OpenVINO-enabled models.