Worked on refactoring offline text-to-speech model asset loading in the alibaba/MNN repository, focusing on improving how model configurations are managed within Sherpa MNN TTS components. Migrated the loading process from using embedded model data to a file-path-based approach, updating initialization workflows and MNN::Express::Module::load calls to support external assets. This change enhanced maintainability and deployment readiness by decoupling model assets from the codebase, making updates and asset auditing more straightforward. The work leveraged C++ and embedded systems expertise, with additional experience in Java and mobile development, and addressed asset management reliability for speech synthesis applications without introducing new bugs.