Worked on the openvinotoolkit/openvino repository to enhance the model import and deployment pipeline by implementing device-aware model import deserialization. Introduced an import_config parameter to the import_models function, enabling propagation of device-capability-specific settings during deserialization of compiled submodels. Leveraged C++ and backend development skills to standardize configuration flow and ensure consistent behavior across NPU devices. Enabled conditional sequencing of inferences when specific NPU settings are active, improving deterministic execution and reliability. Focused on model serialization to reduce production inconsistencies, resulting in a more robust and predictable deployment process for models across diverse hardware environments without fixing any bugs.