Kagura Chen developed security-focused features and robustness improvements for the Hermes Agent and DeepTutor repositories, concentrating on data integrity and reliable parsing. Using Python, Kagura enhanced the Hermes Agent’s chat completions API to filter transcript-only roles and implemented a two-layer defense against sensitive metadata leakage, reducing data exposure risks. Kagura also addressed YAML parsing issues by normalizing server names and adding regression tests to prevent type errors. For DeepTutor, Kagura unified JSON parsing across agent modules to handle LLM outputs wrapped in markdown fences, improving error handling and code clarity. The work demonstrated depth in backend development and API integration.