Architecting, Understanding & Securing Agentic AI systems

My research focuses on designing, understanding, and securing agentic AI systems. AI Agents that plan, use tools, and act in real workflows, by grounding autonomy in data science. I treat trust as something that can be measured and improved using data-driven evaluation and monitoring. My work addresses the automation gap that can emerge in agentic workflows, especially under risks such as prompt/tool injection, unsafe autonomy, and unintended data exposure. To reduce these risks, I develop practical safeguards including least-privilege tool access, policy-based guardrails, and runtime monitoring. My experiments combine agent architectures with evaluation methods that reveal failure modes in multi-agentic workflows, enabling more transparent and controllable executions. Overall, my goal is to enable agentic systems that are not only capable, but also reliable, auditable, and safe to deploy.

Google Scholar Profile

Applied Artificial Intelligence in Healthcare, Deep Learning and Neural Networks, Systems Architecting and Systems Thinking, Many-Objective Optimization

My research applies artificial intelligence and systems engineering methodologies to improve decision-making and operational performance within healthcare systems, with a specific focus on kidney allocation from the Organ Procurement Organization (OPO) perspective. I examine how allocation policies, operational workflows, and stakeholder decisions interact within a complex adaptive healthcare environment to influence system-level outcomes, particularly the trade-offs between utility, fairness, and efficiency (time) in kidney transplantation processes. Using data-driven modeling, simulation, and optimization techniques, I analyze complex interactions among clinical criteria, logistical constraints, stakeholder decisions and policy frameworks that shape kidney allocation decisions and drive emergent system behavior. By viewing the transplant network as a Complex Adaptive System, my work emphasizes how local rules, adaptive stakeholder behaviors, feedback mechanisms and operational practices collectively produce emergent patterns across the broader allocation ecosystem. Through systems thinking and systems architecting, I aim to develop intelligent decision-support tools that help OPOs evaluate policy alternatives, improve transparency, and support equitable and effective allocation strategies. In general, my research seeks to advance policy-informed, scalable, and resilient adaptive healthcare systems that enhance both efficiency and fairness in kidney allocation outcomes while aligning operational performance with ethical and regulatory requirements.

Google Scholar Profile

Scientific Machine Learning, Optimal Control, Fourier Series based Neural Networks

Under the supervision of Dr. Gabriel Nicolosi, my research bridges the mathematical theory of optimal control with scientific machine learning. My goal is to create interpretable, efficient control architectures that enable better decision-making and policy design for nonlinear and uncertain real-world systems.

We developed a novel neural network architecture called Fourier Learning Machines (FLMs), which embed multidimensional nonharmonic Fourier representations directly into a feedforward neural structure. We apply these architectures to numerically solve high-dimensional optimal control problems and partial differential equations, where classical methods often struggle.

We use FLMs to solve the Hamilton-Jacobi-Bellman equation alongside benchmark challenges like the Heston-Merton optimal control problem with stochastic volatility in an incomplete market. In parallel, I am expanding my research into physics-informed machine learning for manufacturing. In collaboration with Missouri S&T’s Materials Science and Engineering department, we apply these advanced techniques to optimize hot-rolling processes.

Overall, my research aims to advance the integration of mathematical control theory and machine learning, developing structured learning systems that can address complex decision-making challenges across finance, engineering, and industrial applications.

Google Scholar Profile 

Systems Engineering

AI in healthcare, implementing digital twins across the entire transplant lifecycle, medical devices

Data Analytics, Queueing Theory, Operations Management

My work focuses on modeling, analysis, and optimization of manufacturing and service systems using queueing theory, simulation, and data-driven techniques. To this end, I am adopting a systems-level approach to generate key insights on the real-world system to be modeled. This involves approaching the system from numerous angles, including approximations. A key goal of my research is to explore the accuracy and practical applicability of these approximations under congestion conditions, commonly observed in real-world production-inventory systems. My work integrates optimization and approximation approaches with simulation models to improve operational decision-making in these systems. In addition, I am interested in benchmarking heuristic approaches from industry, such as drum buffer rope from the lean manufacturing literature. 

Google Scholar Profile

High dimensional Data Analysis

My research focuses on analyzing data in very high-dimensional spaces to improve prediction and classification while ensuring stronger explainability, interpretability, and integrity in artificial intelligence systems. Modern machine learning models often represent data using high-dimensional embeddings; for example, natural language text can be encoded in vector spaces with hundreds of dimensions, such as 768-dimensional representations. These embeddings capture semantic relationships that allow models to better understand complex patterns in data. The techniques I study can be applied across multiple modalities, including genomics, video, audio, images, and natural language.

A fundamental challenge in such environments is the curse of dimensionality, which can reduce the effectiveness of traditional distance measures and learning algorithms. My work investigates approaches to address this challenge by leveraging manifold-aware distance measures and geometric analysis of high-dimensional representations. The goal is to develop models that maintain reliable performance while also providing greater transparency and interpretability.

One of my current research directions focuses on TXT2SQL semantic alignment. In this setting, non-technical users express database queries using natural language. These inputs must be translated into SQL queries that accurately reflect the user’s intent. I analyze both the natural language input and candidate SQL queries in a high-dimensional embedding space, typically around 1536 dimensions, to measure semantic alignment. By detecting misalignment between the intended meaning and the generated SQL query, the system can prevent incorrect database executions that might otherwise produce misleading results. This approach improves the reliability, interpretability, and trustworthiness of AI-assisted database querying systems.

Google Scholar Profile