Kushaan Sharma Kushaan Sharma

Summer Intern, NIH IRTA

National Institutes of Health, NIAAA · Laboratory for Integrative Neuroscience

Built a real-time sleep-scoring pipeline in Open Ephys that classifies EEG and EMG as it streams, across up to 20 animals at once. When the scorer calls NREM, the rig fires a TTL pulse and delivers optical stimulation, closing the loop between what the brain is doing and what the experiment does about it.

The surrounding infrastructure matters as much as the classifier: automated run logging, benchmark testing, and portable run folders, so a night of recording is reproducible rather than artisanal. The system is now in NIH technology transfer for a patent and license.

Machine Learning Research Intern, ML4H

The Broad Institute of MIT and Harvard

Tensorized the EchoNext cohort into 82,543 HD5 records and trained unconditional and cross-attention-conditioned diffusion models for 12-lead ECG synthesis on NCSA Delta A100 nodes, benchmarked against a supervised structural-heart-disease baseline (test ROC AUC 0.797, n = 5,440).

Audited conditioning fidelity by paired sampling from fixed noise: label flips shifted the output less than a 1% noise perturbation (21% vs. 23% of signal RMS) and were largely seed-specific. Traced the cause to an inactive cross-attention control gate and confirmed it by gradient probe. Membership-inference testing showed no evidence of memorization (AUC 0.479).

Health Technician

U.S. Department of Veterans Affairs · San Francisco VA Health Care System

Generated high-quality structured hexahedral meshes from cardiovascular imaging data using TrueGrid, turning patient imaging into the structured geometry that computational simulation of blood flow and vessel mechanics requires.

Research Assistant and Project Lead, Pu Laboratory

Department of Cardiology, Boston Children's Hospital · Harvard Medical School

Jul to Aug 2026. Evaluated computational methods for 2D analysis and 3D reconstruction of serial MERFISH sections from early-stage embryos: benchmarked cell-segmentation approaches (membrane markers, DAPI, transcript density) and compared tools for section alignment, spatial transcriptomics integration, and 3D visualization.

Jan to Aug 2026, project lead. Developed a Balanced Random Forest classifier that detects atrial fibrillation in murine ECGs, and built an open-source annotated mouse ECG library for the wider research community.

Sep 2025 to Jan 2026. Built the hardware-software integration for engineered heart tissue analysis, writing MATLAB signal-processing algorithms from scratch to measure contractility, deflection, and pacing response. Released as EHT-analyze and co-authored on a Nature Protocols manuscript now in revision. The work earned a merit-based research stipend.

Project Lead, Virostko Laboratory

Dell Medical School and Oden Institute · UT Austin

Built two 3D MRI analysis pipelines: one quantifies pancreatic fat as a function of distance from the organ surface, the other separates intralobular, interlobular, and peripancreatic fat depots. Together they showed that spatial heterogeneity explains a two-fold discrepancy between the two standard measurement methods, and that compartment-level measures detect diet-induced change invisible to conventional analysis.

First-author manuscript under review at Investigative Radiology; abstract accepted at the 2026 American Pancreatic Association Annual Meeting.

Research Assistant, Nanoassembly Lab

Department of Biomedical Engineering · Cockrell School of Engineering, UT Austin

Analyzed how the N/P ratio affects the stability of DNA origami nanostructures, using qPCR degradation measurements to quantify how much intact structure survives under nuclease challenge. Listed on the resulting manuscript.