Responsible AI for Healthcare.
into actionable clinical insights.
impracticality.
Focused Experience
Graduate Research Assistant
Shipman Lab & Erraguntla Lab, Texas A&M
- ▹Performed K-fold cross-validation over an ensemble network (using VAE and 12-lead ECG spectrograms) achieving AUC > 90% for cardiovascular classification.
- ▹Utilized LIME analysis for explainability and engineered an LLM integration for doctor's notes, improving real-time multi-modal predictions.
- ▹Built an ECG-to-text model leveraging LLMs, trained on ~50,000 samples for automated text-based clinical interpretations.
- ▹Led ML research for a $1M FDA-funded project on infant Aortopulmonary Shunt occlusion, architecting an XGBoost ensemble scoring peak AUC 0.867.
- ▹Implemented a multi-layered XAI framework using SHAP to generate actionable clinical narratives for PICU teams.
Data Science Intern
Delfina
- ▹Developed a Gradient Boosting Classifier predicting Gestational Weight Gain via clinical EHR and self-reported data across 4 clinics (AUC 0.85).
- ▹Engineered longitudinal weight trajectory features combined with multi-modal predictors such as BMI and parity.
- ▹Conducted rigorous bias/stability analysis utilizing DeLong’s test, proving model stability (p > 0.05) across diverse racial and BMI categories.
Graduate Research Assistant
Perception, Sensing, and Instrumentation Lab, Texas A&M
- ▹Engineered CNN and LSTM models to analyze ECG morphology and HRV features to achieve multi-threshold glycemic classification (hypo, normo, hyper).
- ▹Led protocol development of multi-modal physiological data for 12 T1D subjects over a 14-day tracking period.
- ▹Designed a custom 3-way classification loss function for federated learning, preserving privacy for 120 patients.
Clinical Research Intern
Edwards Life Sciences
- ▹Designed an ensemble deep learning model to automate heart murmur detection from phonocardiograms using Mel spectrograms and YAM Net.
- ▹Developed interpretable ML models (AdaBoost, XGBoost) for pacemaker complication prediction.
- ▹Constructed robust automated PyCaret pipelines using bootstrap sampling to validate predictive capabilities for cardiac care support.
AI Research Intern
Philips
- ▹Engineered an IoT-driven AI pipeline using GNNs and Raspberry Pi for real-time anomaly detection in veterans' utility usage.
- ▹Developed an infrared (IR) CNN-based eye-temperature detection system to perform real-time screening at building entrances.
Academic Projects
SNOOZE: LLM powered intelligent organizer
Engineered an intelligent email organizer leveraging LangChain, agentic LLMs, and a graph-based LangGraph structure for automated email management, scheduling, summarization, and task creation. Integrated Google API management and authentication protocols.
Echoes of Choices: Behavior and stress analysis
Developed an immersive, decision-based narrative game powered by LLMs that dynamically adapts storylines based on user choices. Created a custom dataset of 50 text-based stories incorporating behavior metrics and psychological evaluation to generate personalized reports.
Hybrid Branch Predictor
Designed a high-performance branch predictor using a perceptron-based hybrid selection mechanism to meet industry benchmarks, optimizing machine-level code efficiency and predictive accuracy.
Skills & Competencies
Machine Learning
Deep Learning, CNN, Time series inference, NLP, RNN, LLM, Generative AI, interpretability, unsupervised learning
Languages & Frameworks
Python (Keras, PyTorch, TensorFlow), Langchain, R, C++
Databases
MySQL, PostgreSQL, MongoDB
Datatypes
Physiology signals, electronic health records, medical images, Log data, point cloud
Education
Texas A&M University
PhD in Computer Science, Expected May 2026
Dissertation: Bridging The Valley Of Impracticality: A Progressive Disclosure And XAI Framework For Actionable Clinical Decision Review System
Northeastern University
Master of Science in Electrical and Computer Engineering (May 2021)
Master of Professional Studies in Informatics - Data Science track (May 2018)
Selected Accolades
Key Publications & Patents
- Patent: "Multi-sensor upper arm band for physiological measurements... predict hypoglycemia/hyperglycemia"
- Vyas, K. et al. (2019). Recognition of atypical behavior in autism diagnosis from video. IEEE MLSP.
- Dave, D., Vyas, K. et al. (2024). FedGlu: A personalized federated learning-based glucose forecasting algorithm.
- Vyas, K. et al. (2020). Additional value of augmenting current subscales in braden scale... for pressure injury risk assessment. IEEE BIBM.
Honors & Awards
- Aggie Core Values Award 2025
- Graduate Leadership Excellence Award 2025
- Buck Weirus Spirit Award 2025
- ASIE Scholarship Recipient 2024