Worked on backend and server development for IBM/vllm and rmusser01/llama.cpp, focusing on stability, reliability, and resource optimization. Addressed a critical illegal memory access bug in IBM/vllm by refining memory handling when advanced features like chunked prefill and xformers were enabled, and introduced regression tests to ensure robust model behavior across diverse prompt configurations. In rmusser01/llama.cpp, implemented an LCS-based server slot allocation algorithm in C++ to improve task-slot matching and resource utilization, and enhanced scheduling reliability with smarter slot selection strategies. Leveraged C++, Python, and algorithm optimization skills to deliver targeted improvements in machine learning infrastructure.