Aryaman Velampalli
I am an MS Computer Science student on the ML track at Columbia University, graduating December 2026. I completed my B.E. in Computer Science from BITS Pilani in 2023.
I do research in Professor Mohammed AlQuraishi's computational biology lab, where I work with protein backbone diffusion models. Previously, I spent two years as an ML Engineer at Avalara, where I built a conversational Taxability Research Assistant across 3M+ tax documents.
My research interests lie in diffusion models, flow matching and reinforcement learning and their applications to problems in biology. I also try to keep up with the rapidly changing landscape of large language models.
Projects
-
A generative model using diffusion principles, transformer and reinforcement learning to design 3D protein-like chain structures. Trains a transformer-based DDPM/DDIM denoiser and fine-tunes it with DDPO policy gradients to optimize structural properties such as beta-sheet alignment. -
An implementation of DDPM and DDIM on 2D toy datasets (circles, two-moons, spirals) using an MLP denoiser with sinusoidal embeddings and cosine noise scheduling. Built to develop intuition for the forward and reverse diffusion process through clean visualizations.