Developed the InsertPunctuationConverter feature for the Azure/PyRIT repository, focusing on enhancing prompt robustness in large language models. This module enables configurable insertion of punctuation either between or within words, allowing for adjustable perturbation ratios to stress-test prompt handling. The implementation leveraged Python and Jinja, with comprehensive unit tests designed to validate the module’s effectiveness across diverse scenarios. By linking this work to the Moonshot Attack Module, the developer enabled measurable evaluation of prompt perturbation impacts, supporting safer deployment of LLM-driven features. The project emphasized natural language processing, prompt engineering, and rigorous software testing to improve model resilience and data integrity.