Over a two-month period, contributed to both leanprover-community/mathlib4 and inclusionAI/AReaL by delivering targeted feature enhancements. In mathlib4, developed and formalized new lemmas in Lean for digit-based reasoning in number theory, establishing an equivalence between the first i digits of a number in base b and modulo base powers, which improved modular reasoning and aligned with existing tooling. For AReaL, implemented a Triton kernel in Python to optimize tree attention computations, introducing conditional logic for efficient resource use and improved memory management. Work demonstrated depth in formal verification, CUDA, and machine learning, focusing on robust, maintainable feature delivery.