Yuhan Guo contributed to the pytorch/torchchat and pytorch/executorch repositories by developing features that enhance deployment reliability and metadata visibility. In torchchat, Yuhan improved installation workflows using Shell scripting and Python, introducing robust error handling and embedding tokenizer type information into exported models to streamline deployment and reduce runtime issues. For executorch, Yuhan implemented a C++ feature that enables retrieval of method attributes within the training module, improving traceability and supporting future metadata-driven analytics. The work demonstrated a thoughtful approach to software architecture and unit testing, delivering targeted improvements that address deployment friction and facilitate more transparent model training and evaluation.