About

I am Amit Subhash. I build things, constantly. I am an incoming PhD student in Neuroengineering at Indiana University, where I work on Diffuse Optical Tomography (DOT) for neonatal brain imaging. I previously completed an MS in Data Science.

By the numbers

What I work on

My research centers on Diffuse Optical Tomography — a non-ionizing, low-cost neuroimaging modality that uses near-infrared light to reconstruct hemodynamic activity in the brain. DOT is uniquely suited to neonatal brain imaging: portable, safe at the bedside, and effective through the thin neonatal skull. The hard problems are the forward model (Monte Carlo photon transport in heterogeneous tissue), the inverse problem (ill-posed reconstruction from sparse boundary measurements), and real-time performance on commodity hardware.

I attack each of these with a different stack. Classical regularization (Tikhonov, GCV) and modern learned priors (diffusion posterior sampling, score-based generative models) for the inverse side. Differentiable Monte Carlo for end-to-end optimization through the forward model. SynthSeg-style domain randomization plus SAM2 for contrast-agnostic anatomical priors. Sparse autoencoders for understanding what tiny biosignal transformers actually learn. And underneath all of it, an opinionated research infrastructure of arXiv watchers, DSPy classifiers, and HPC reference docs that turn a one-person lab into something that ships.

Stack

Affiliations

How I work

I prefer single long sessions. I research and reuse before building net-new — proven implementations beat novelty almost every time. I write conventional commits, run TDD on new code, and review before merging. I push almost every day, because research compounds when the loop is tight.

Contact