
0
0
India
2+ years of experience
Optimizing AI Models (LLMs) at Harvard Reduced LLM inference time by 20% by optimizing VIDUR’s control knobs. Integrated Meta-Llama and CodeLlama into a framework, improving tool usage by 15%. Generated 10,000+ synthetic datasets for testing and improved distributed computing performance. AI for Medical Imaging at JuliaHealth (Julia Lang) Implemented BlochKernel, a GPU-accelerated simulation method, improving medical imaging speed by 40%. Enhanced GPU simulations across three architectures, reducing compute time by 35%. Automated pipelines on Buildkite, cutting testing time by 50%. AI for Biomarker Discovery at NIH (USA Govt) Analyzed 100+ gene expression datasets for lung cancer biomarker detection. Achieved 90% accuracy for serum samples and 99% for plasma samples. Used statistical methods (Bartlett's, Levene's, ANOVA) for validation and optimized pipelines, reducing data processing time by 30%.