Worked on the jeejeelee/vllm repository to enhance the reliability of multimodal generation by upgrading the Nemotron Parse testing framework to version 1.2. Focused on AI model integration and machine learning, the developer re-enabled the parity test, aligning validation processes with the updated nemotron-parse model. This approach improved test accuracy and restored comprehensive parity coverage in continuous integration, reducing the risk of regressions in production. The work was implemented in Python and emphasized robust testing practices, ensuring traceability and compatibility with CI pipelines. The contribution centered on strengthening test coverage rather than bug fixes, reflecting a methodical and quality-driven approach.