Worked on the ggml-org/llama.cpp repository to enhance the stability and reliability of its command-line interface, focusing on parameter handling for model execution. Addressed a critical issue where the default sampling parameter was not correctly derived from user-provided inputs, which previously led to configuration errors and inconsistent behavior across different model settings. By implementing a targeted bug fix in C++, ensured that the CLI now consistently propagates parameters, improving reproducibility and deployment reliability for users. The work demonstrated strong command line interface development skills and a methodical approach to debugging, resulting in more predictable and robust model runs for the project.