Archana Ramalingam enhanced the nod-ai/SHARK-Platform by improving perplexity evaluation reliability and streamlining continuous integration workflows. She addressed stability issues in the Perplexity class by refining sequence length handling and cache state access, ensuring more accurate model evaluation. Using Python and PyTorch, Archana also updated CI pipelines to trigger on pull requests, adjusted test parameters, and improved logging for better observability. Her work included reverting and tuning previous CI changes to optimize test stability, introducing new test prompts, and switching device targets for perplexity tests. These targeted engineering efforts resulted in faster feedback cycles and higher quality model assessment.