CV

PhD in Informatics (Health & Biomedical Informatics)

• Advisor: Dr. Saptarshi Purkayastha, Purkayastha Lab for Health Innovation
• Research focus on multimodal clinical AI, medical image analysis, and clinical decision support

Bachelor of Science in Electrical Engineering

• Concentration in Machine Learning and Controls
• Research Assistant in Statistical Visual Computing Lab, advised by Prof. Nuno Vasconcelos
• Research Assistant in Xu Research Group, advised by Prof. Sheng Xu

Tri-Modal Fusion for Parkinson's Disease Motor Phenotype Classification

• Developed a hierarchical multimodal pipeline to classify 130 participants into Healthy Control, Tremor-Dominant PD, Intermediate PD, and PIGD PD using resting EEG, walking EEG, and clinical features derived from MDS-UPDRS Part III TD/PIGD ratios.
• Built a two-stage architecture combining an XGBoost HC-vs-PD gate, EEGNet-Transformer rest/walk encoders, bidirectional EEG cross-attention, and clinical XGBoost soft fusion for within-PD subtype classification.
• Improved 4-class LOGO macro-F1 by approximately +0.10 over flat trimodal cross-attention; validated with LOSO/LOGO evaluation, multi-seed robustness testing, bootstrap 95% CIs, and Wilcoxon signed-rank tests.
• Achieved 0.56 ± 0.02 macro-F1, 0.98 ± 0.01 HC-vs-PD gate AUC, and 0.43 ± 0.02 PD-only 3-class F1; used SHAP TreeExplainer to identify Color Trails Test and MoCA as dominant within-PD predictors.

TriageGuard: LLM-Driven Continuous Security Testing for Open-Source EHRs

• Developed TriageGuard, a PR-triggered OpenMRS security-testing pipeline that converts code diffs into executable Gherkin/Playwright tests and validates them against a Dockerized OpenMRS instance through GitHub Actions CI.
• Built a dual-LLM safety gate using Qwen-3-32B generation, GPT-OSS-120B review, SBERT semantic verification, and 13-condition approval logic to block unsafe AI-generated tests before execution.
• Integrated CodeQL static analysis with 4 custom queries covering unprotected methods, privilege annotations, HTTP entry points, and caller chains; grounded test generation with an 8-profile Gherkin scenario library.
• Operationalized CVSS 4.0 risk-band scoring and a 4-signal composite risk model combining evidence density, OWASP pattern matching, CodeQL corroboration, and model self-reporting, with SHA-256 provenance for every LLM call.

Multimodal EEG-IMU Fusion for Robust Motor Assessment

• Developed a multimodal motor-assessment framework integrating 16-channel EEG and wrist-worn IMU signals across 10 motor activities to evaluate neural and kinematic modality complementarity.
• Implemented an EEGNet-Transformer model (~230K parameters) and a feature-engineered XGBoost IMU pipeline using 76 features per sensor across time-domain, frequency-domain, and 3D composite groups.
• Achieved 98.68 ± 0.32% activity-classification accuracy through logistic-regression late fusion, reducing EEG-only error by 81.5% and improving worst-task accuracy from ~87% to 96.76%.
• First-author paper accepted and presented at IEEE ICHI 2026.

Human-Guided Agentic AI for Multimodal Clinical Prediction

• Contributed to an AgentDS Healthcare benchmark submission; placed 5th overall in the healthcare domain and 3rd on discharge-readiness prediction.
• Led 30-day readmission prediction across 5,000 admissions, engineering 871 structured-EHR and discharge-note features from admission records, demographics, 5-year ED cost history, and trigram TF-IDF notes.
• Built a 5-base-model stacking ensemble using 3 XGBoost variants, L1/L2 logistic regression, and 8-fold out-of-fold meta-learning; achieved test Macro-F1 = 0.8986, within 0.006 of the top-ranked system.
• Co-author paper accepted and presented at IEEE ICHI 2026.

Frozen Foundation-Model Embeddings for Small-Lesion CXR Analysis

• Co-investigated preservation of small-scale diagnostic signal in frozen ViT foundation-model embeddings for chest radiography across 5 frozen models evaluated on NIH-CXR14, MIMIC-CXR, Emory-CXR, and ChestX-Det10 cohorts.
• Evaluated 492,724 chest radiographs and 3,543 ChestX-Det10 images with 1,462 lesion boxes across calcification, nodule, and mass categories.
• Demonstrated CLS pooling collapses small-scale signal to chance-level AUC = 0.500–0.524, while ROI-conditioned patch tokens recover ~AUC = 1.0 with mean per-model gains of +0.412 to +0.488.
• Second-author manuscript under review at Medical Image Analysis.

Summer Research Assistant

• Advised by Dr. Bridget Bannerman & Prof. Jorge Júlvez
• Conducted systems biology modeling to investigate how Mycobacterium tuberculosis infection alters macrophage metabolism and contributes to lung cancer progression.
• Implemented genome-scale metabolic models (iAB-AMØ-1410, iAB-AMØ-1410-Mt-661) in COBRApy to perform Flux Balance and Flux Variability Analyses for host–pathogen interaction mapping.
• Identified 223 infection-specific essential reactions and quantified nutrient dependencies through medium minimization to reveal candidate metabolic drug targets.
• Simulated infection-induced metabolic burden, showing a 92% reduction in macrophage biomass growth rate due to Mtb-driven energetic constraints.

Undergraduate Research Assistant

• Advised by Prof. Maurício de Oliveira & Prof. Jack Silberman
• Built a deep-learning autonomous vehicle on Raspberry Pi using TensorFlow, OpenCV, and the DonkeyCar framework for real-time lane detection, steering prediction, and obstacle avoidance with sub-150 ms inference latency.
• Curated and trained CNNs on 35,000+ labeled image frames while integrating sensors, motor drivers, and electronic control modules.
• Reduced vehicle failure rate from 25% to 3% through iterative testing and model optimization, enabling fully autonomous operation across 10 indoor and 6 outdoor laps.

Undergraduate Research Assistant

• Advised by Prof. Nuno Vasconcelos
• Developed a deep learning-based system for automated plankton image labeling and retrieval to support ecological research at the Scripps Institution of Oceanography.
• Applied transfer learning with ResNet-50 and InceptionResNetV2 and integrated Approximate Nearest Neighbor (ANN) search for large-scale image similarity retrieval.
• Boosted fine-grained plankton classification accuracy to 90% through targeted data augmentation strategies.

Undergraduate Research Assistant

• Advised by Prof. Sheng Xu
• Fabricated a stretchable conformal ultrasonic probe using laser micro-cutting and soft elastomer encapsulation for non-invasive cardiovascular monitoring.
• Modeled ultrasound beam patterns and penetration depth in MATLAB, validating simulation results against experimental measurements.
• Evaluated device stability and acoustic coupling under mechanical deformation to support signal fidelity during motion.
• Prepared IRB documentation and risk assessments for human-subject testing.
• Contributed to the resulting Nature Biomedical Engineering publication (2018).

Healthcare IT Solution Team Lead

• Led 5+ concurrent healthcare IT implementation projects across Yunnan Province, coordinating engineering teams, hospital administrators, and third-party vendors.
• Directed EHR system integration for client hospitals, aligning data interoperability, implementation continuity, and clinical workflow requirements.
• Improved project delivery rate from 75% to 90% by introducing milestone tracking, risk forecasting, and performance-monitoring workflows.
• Supported design and deployment of regional telemedicine and Internet Hospital platforms on China Mobile Cloud, reducing onboarding costs for smaller healthcare institutions by 55%.

Healthcare IT Project Solution Manager

• Managed end-to-end delivery of large-scale healthcare digital transformation projects totaling 21 million CNY, spanning medical networks, healthcare data platforms, and clinical information systems.
• Designed the VPDN Medical Insurance Account Management System; produced technical specifications and functional prototypes that improved delivery efficiency by 15% and achieved national software copyright certification.
• Led deployment of the Fangcang Shelter Hospital medical data transmission network during the COVID-19 response, completing implementation 10 days ahead of schedule while reducing labor costs by 8%.
• Collaborated with hospitals, insurers, and telecom engineering teams to deliver secure, high-availability medical data transmission infrastructure.

• Yin Z, Pulavarthy LP, Purkayastha S. "Multimodal EEG-IMU Fusion for Motor Assessment: Leveraging Task-Dependent Complementarity for Robustness." IEEE International Conference on Healthcare Informatics (ICHI), 2026.
• Pulavarthy LP, Muthyala R, Kuruvikkattil AV, Yin Z, Kudamala R, Purkayastha S. "Human-Guided Agentic AI for Multimodal Clinical Prediction: Lessons from the AgentDS Healthcare Benchmark." IEEE ICHI, 2026.

• Wang C, Li X, Hu H, Zhang L, Huang Z, Lin M, Zhang Z, Yin Z, et al. "Monitoring of the central blood pressure waveform via a conformal ultrasonic device." Nature Biomedical Engineering, 2(9), 687-695. doi 10.1038/s41551-018-0287-x

• China Mobile VPDN Management System. National Software Copyright Registration, China. Certificate No. 10834368; Registration No. 2023SR0247197.

• Oral Presentation, First Author. "Multimodal EEG-IMU Fusion for Motor Assessment: Leveraging Task-Dependent Complementarity for Robustness." IEEE International Conference on Healthcare Informatics (ICHI), Minneapolis, MN, Jun 2026.
• Oral Presentation, Co-Author. "Human-Guided Agentic AI for Multimodal Clinical Prediction: Lessons from the AgentDS Healthcare Benchmark." Data Challenge Program @ IEEE ICHI 2026, Minneapolis, MN, Jun 2026.

• Oral Presentation, Co-Author. "Construction of a Hospital-wide Critical Illness Rescue Network Based on 5G for Severe Illness Early Warning Platform." 30 Finalists, 5G+ Healthcare Special Competition Finals, The 7th "Blooming Cup" 5G Application Competition.

• University Fellowship, Indiana University
• Human Research Biomedical Researcher - Stage 1, CITI
• Biomedical Responsible Conduct of Research, CITI
• GCP - Social and Behavioral Research Best Practices for Clinical Research, CITI

• Qualification of Telecommunication Professional, Ministry of Human Resources and Social Security & Ministry of Industry and Information Technology, China

• Certified LabVIEW Associate Developer (CLAD), National Instruments