Worked on the NVIDIA/TensorRT-LLM repository to enhance context management flexibility by refactoring the mContextChunkSize configuration. This change introduced separate fields for target and draft models, allowing independent tuning of context chunk sizes and reducing global coupling within the codebase. The approach focused on improving maintainability and enabling more precise experimentation with context length for varied workloads. Utilizing C++ for development, the work emphasized performance optimization and thoughtful software architecture. The resulting feature supports better performance tuning opportunities and prepares the TensorRT-LLM integration for future enhancements, reflecting a methodical approach to scalable and adaptable system design.