During a two-month period, Alumkal contributed to leanprover-community/mathlib4 and inclusionAI/AReaL, focusing on both formal verification and deep learning. In mathlib4, Alumkal formalized digit-based lemmas in number theory, establishing the equivalence between taking the first n digits of a number in base b and performing a modulo operation, using Lean and number theory expertise to enhance modular reasoning tools. For AReaL, Alumkal implemented a Triton kernel for tree attention, introducing conditional logic to optimize memory and computation for large tree-structured data. The work demonstrated depth in both formal methods and high-performance CUDA-based machine learning engineering.