Worked on the tenstorrent/tt-inference-server repository to enhance the performance reference JSON used for benchmarking multiple models. Focused on updating and reformatting the JSON structure, the work involved adding missing benchmarking targets and improving data organization for better readability and usability. Leveraged skills in JSON manipulation, data formatting, and performance benchmarking to ensure the reference data supports reliable cross-model comparisons and seamless integration with testing pipelines. These improvements enable faster evaluation and decision-making during model deployment, while also streamlining onboarding and continuous integration processes. The changes were delivered as a single feature update, with traceable, well-documented commits.