Peer-reviewed work spanning AI security, privacy, medical devices, public safety, and sustainable computing.
UAB Intellectual Property Disclosure, 2025. Embeds IoT sensors and computing capability directly into building materials to enable smart, sustainable architectural infrastructure with minimal added cost.
Poster presentation at AI in Nephrology Conference 2026 showcasing the CRRT-MAPPER which match CRRT machine log data with the corresponding patient health records at scale
Poster presentation IEEE ICCC illustrating The integration of fuzzy logic with interaction provenance by enabling dynamic, context-aware access decisions based on uncertainty, historical interactions, and behavioral evidence, while ensuring improved security, flexibility, and traceability.
A proposed structural authorization layer, utilizing signed tokens and a policy oracle, prevents multi-agent LLM systems from executing unauthorized actions, even when upstream validation fails.
PDFxRay is a RAG-enhanced LLM framework that secures hospital networks by analyzing malicious healthcare PDFs, identifying threat patterns, and providing actionable, explainable risk assessments.
TwinGuard simulates hospital employee access privileges using digital twins, evaluating graph-based attack paths and risk scores to block cyber threats and protect medical networks before new accounts are activated.
LabOrchestrator is an AI-driven framework that unifies automated threat modeling, explainable access control, and continuous compliance monitoring to provide adaptive, end-to-end cybersecurity for medical research environments.
ComplianceGPT is an LLM-based, context-aware agent that automatically detects medical data privacy risks, explains regulatory violations, and provides actionable de-identification guidance to support healthcare compliance.
LLMAC unifies traditional access control models using Large Language Models to deliver highly accurate, context-aware, and explainable access decisions for complex, dynamic organizational environments.
IPBAC integrates fuzzy logic with interaction provenance by enabling dynamic, context-aware access decisions based on uncertainty, historical interactions, and behavioral evidence, while ensuring improved security, flexibility, and traceability.
A secure peer-to-peer firmware update protocol for drones and autonomous aerial vehicles, preventing malicious update injection without a centralized authority while balancing real-world bandwidth and latency constraints.
SmartWall is an intelligent, low-cost IoT-enabled building component that provides real-time structural and environmental monitoring to enhance public safety, resilience, and sustainability in smart cities.
This study demonstrates that transformer-based models significantly outperform traditional machine learning methods in classifying disaster-related tweets, enabling more accurate and context-aware public safety monitoring.
An agentic LLM pipeline automating STRIDE-style threat modeling for public-safety infrastructure — identifying attack surfaces, generating threat scenarios, and recommending mitigations.
Maps realistic adversarial attack paths against network-connected radiological devices (MRI, CT, X-ray) using MITRE ATT&CK, identifying critical vulnerabilities and proposing mitigations aligned with healthcare security standards.
This research introduces a graph-based machine learning approach for accurate and efficient detection of vulnerabilities in Ethereum smart contracts.
Benchmarks energy consumption across diverse ML architectures during training and inference, introducing a sustainability metric to guide eco-conscious model selection without sacrificing predictive accuracy.
Uses belief rule base reasoning to compare and rank acoustic features from speech signals for Parkinson's diagnosis, improving diagnostic confidence by modeling the inherent uncertainty in neurological symptom expression.
Combines CNN feature extraction with a belief rule-based expert system to classify visual sentiment from images under uncertain labeling, outperforming standalone deep learning on ambiguous datasets.
Fuses an epidemiological SEIR model with a neural network to predict COVID-19 trajectories, capturing both disease dynamics and data-driven nonlinear patterns for higher forecast accuracy than either approach alone.
Applies CNN-based spectral feature learning to classify emotional states from speech of patients with neurological disorders, outperforming traditional handcrafted feature pipelines on pathological speech data.