Kumarran Mahesh

I’m a third-year B.Tech (Honors) student in Artificial Intelligence & Data Science, focused on building applied machine-learning systems end to end—retrieval, evaluation harnesses, vision pipelines, and small production services. I’m especially drawn to work that connects model quality with practical deployment.

Currently, I’m a Machine Learning Research Collaborator with the University of Rochester, where I built a 33M-parameter residual attention U-Net with a learnable spectral encoder for multispectral image restoration. In my projects, I shipped an end-to-end Customer Segmentation & CLV platform with stable clustering quality, and I built a high-throughput RAG framework using FAISS and a streaming ingestion pipeline over the 300GB Hugging Face C4 corpus. I also developed a degradation-aware perception pipeline that routes restoration and improves PSNR on low-light inputs.