Research

HAX Lab

Human-AI eXperience Lab, Morgan State University

  • I help design and build intelligent systems that are multimodal and interactive, taking in more than one kind of input and responding to the person using them.
  • I evaluate how people actually experience these systems, testing whether they are usable and whether people are willing to accept them in their daily lives.
  • I study how AI affects the people who rely on it, with the goal of making sure these systems respect human values, not just perform well.

SAIRI Research Internship

CEAMLS Center, Morgan State University · May – Jul 2026

  • Engineered a multimodal concept-drift detection pipeline for early cardiovascular disease prediction, trained across ECG and PPG signals in PyTorch.
  • Built a three-branch architecture: a 1D CNN reading the time domain, a 2D CNN reading time-frequency representations, and a Transformer reading frequency-domain patterns. It reached 87.86% drift-classification accuracy with a 0.8446 energy-drift correlation.
  • Built a cross-modal Complementary Feature Domain (CFD) preprocessing pipeline with pyCFD 3.0, NumPy, Pandas, and scikit-learn, processing over 30,000 physiological signals across PTB-XL (21,000 records) and MIMIC (10,000 segments).
  • Ran four benchmarking experiments on three datasets, then validated externally on the held-out MIMIC waveform set, beating ADWIN, Page-Hinkley, MMD, and Dropout Uncertainty baselines by up to 40 percentage points.
Presentation slide introducing the SAIRI project: an AI pipeline that fuses ECG, PPG, and respiration signals to detect cardiovascular disease risk earlier.
Presenting the SAIRI project introduction at Morgan State.
Team presenting SAIRI drift-classification results, including a PTB-XL heatmap, at a podium branded with the Morgan State seal.
Presenting drift-classification results on the PTB-XL dataset.

© 2026 Rochak Ghimire