About

Researcher and educator interested in sociotechnical systems, machine learning for engineering design, and human-AI teaming

Research topics

• Computer Science
• Artificial Intelligence
• Engineering
• Knowledge management
• Data science
• Human–computer interaction
• Psychology
• Engineering management
• Systems engineering
• Computer Security
• Chemistry
• Applied psychology
• Virology
• Risk analysis (engineering)
• Social psychology
• Business
• Medicine
• Pedagogy
• Management science
• Mathematics education

Selected publications

• Supervising Ralph Wiggum: Exploring a Metacognitive Co-Regulation Agentic AI Loop for Engineering Design

  arXiv (Cornell University) · 2026-03-25

  preprintOpen accessSenior author

  The engineering design research community has studied agentic AI systems that use Large Language Model (LLM) agents to automate the engineering design process. However, these systems are prone to some of the same pathologies that plague humans. Just as human designers, LLM design agents can fixate on existing paradigms and fail to explore alternatives when solving design challenges, potentially leading to suboptimal solutions. In this work, we propose (1) a novel Self-Regulation Loop (SRL), in which the Design Agent self-regulates and explicitly monitors its own metacognition, and (2) a novel Co-Regulation Design Agentic Loop (CRDAL), in which a Metacognitive Co-Regulation Agent assists the Design Agent in metacognition to mitigate design fixation, thereby improving system performance for engineering design tasks. In the battery pack design problem examined here, we found that the novel SRL and CRDAL systems generate designs with better performance, without significantly increasing the computational cost, compared to a plain Ralph Wiggum Loop (RWL) Further, the novel CRDAL generates designs with significantly better performance than SRL. Also, we found that the CRDAL system navigated through the latent design space more effectively than both SRL and RWL. The proposed system architectures and findings of this work provide practical implications for future development of agentic AI systems for engineering design.

  PublisherDOI

• Supervising Ralph Wiggum: Exploring a Metacognitive Co-Regulation Agentic AI Loop for Engineering Design

  ArXiv.org · 2026-03-25

  articleOpen accessSenior author

  The engineering design research community has studied agentic AI systems that use Large Language Model (LLM) agents to automate the engineering design process. However, these systems are prone to some of the same pathologies that plague humans. Just as human designers, LLM design agents can fixate on existing paradigms and fail to explore alternatives when solving design challenges, potentially leading to suboptimal solutions. In this work, we propose (1) a novel Self-Regulation Loop (SRL), in which the Design Agent self-regulates and explicitly monitors its own metacognition, and (2) a novel Co-Regulation Design Agentic Loop (CRDAL), in which a Metacognitive Co-Regulation Agent assists the Design Agent in metacognition to mitigate design fixation, thereby improving system performance for engineering design tasks. In the battery pack design problem examined here, we found that the novel SRL and CRDAL systems generate designs with better performance, without significantly increasing the computational cost, compared to a plain Ralph Wiggum Loop (RWL) Further, the novel CRDAL generates designs with significantly better performance than SRL. Also, we found that the CRDAL system navigated through the latent design space more effectively than both SRL and RWL. The proposed system architectures and findings of this work provide practical implications for future development of agentic AI systems for engineering design.

  PublisherOA PDF

• Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

  Open MIND · 2026-01-20

  preprintSenior author

  Anticipating human activities and their durations is essential in applications such as smart-home automation, simulation-based architectural and urban design, activity-based transportation system simulation, and human-robot collaboration, where adaptive systems must respond to human activities. Existing data-driven agent-based models--from rule-based to deep learning--struggle in low-data environments, limiting their practicality. This paper investigates whether large language models, pre-trained on broad human knowledge, can fill this gap by reasoning about everyday activities from compact contextual cues. We adopt a retrieval-augmented prompting strategy that integrates four sources of context--temporal, spatial, behavioral history, and persona--and evaluate it on the CASAS Aruba smart-home dataset. The evaluation spans two complementary tasks: next-activity prediction with duration estimation, and multi-step daily sequence generation, each tested with various numbers of few-shot examples provided in the prompt. Analyzing few-shot effects reveals how much contextual supervision is sufficient to balance data efficiency and predictive accuracy, particularly in low-data environments. Results show that large language models exhibit strong inherent temporal understanding of human behavior: even in zero-shot settings, they produce coherent daily activity predictions, while adding one or two demonstrations further refines duration calibration and categorical accuracy. Beyond a few examples, performance saturates, indicating diminishing returns. Sequence-level evaluation confirms consistent temporal alignment across few-shot conditions. These findings suggest that pre-trained language models can serve as promising temporal reasoners, capturing both recurring routines and context-dependent behavioral variations, thereby strengthening the behavioral modules of agent-based models.

  DOI

• Energy-based feature extraction with adaptive local domain decomposition for prediction of transient and turbulence flow with operator regression models

  Computers & Fluids · 2026-01-08

  article
  PublisherDOI

• Forging EMPIRE: A data-driven agent-based model for scenario-based generalizability in spatio-temporal human behavior modeling

  Computers Environment and Urban Systems · 2026-02-26

  articleOpen accessSenior author

  People spend the majority of their lives within built environments, whose design can profoundly influence human- and community-centered outcomes such as social capital formation, access to opportunity, public health, and resilience to disruption. Just as the built environment shapes human behavior and well-being, its design, operation, and performance can be substantially improved by better understanding how people actually use and experience space. Yet both of these goals — enhancing human benefits from built environments and improving system performance through human-aware design — are constrained by a fundamental limitation: existing computational models oversimplify human agents, equipping them with static or assumed behavioral rules that fail to reflect the dynamic, adaptive, and context-sensitive nature of real-world behavior. These simplifications undermine generalizability, limiting the ability of such models to transfer insights across scenarios or support the design of responsive, human-centered spaces. To overcome these limitations, we introduce EMPIRE ( Empirical Modeling of People in Responsive Environments ) — a data-driven, hierarchical model for predicting human spatio-temporal behavior in dynamic physical environments, with a focus on scenario-based generalizability. Driven by in-situ data, EMPIRE integrates Imitation Learning for strategic activity planning and Reinforcement Learning for generating adaptive execution policies based on interpretation of the environment and preferences. This multi-layered decomposition mirrors the cognitive structure of human decision making, enabling modularity, interpretability, and adaptability across unseen spatial configurations. To illustrate EMPIRE’s generalizability, we simulate human behavior in a social infrastructure setting (i.e., a park) by generating synthetic ground-truth trajectories that incorporate heterogeneous agent preferences, environmental dynamics, and social constraints. We conduct a systematic evaluation across six distinct park layouts using a leave-one-layout-out strategy, where models are trained on five configurations and tested on the sixth. This setup allows assessment of EMPIRE’s capacity to generalize to various unseen spatial scenarios. Experimental results demonstrate that EMPIRE successfully transfers learned behavioral patterns to new environments. • Data-driven agent-based model learns activities and preferences from in-situ data. • Hierarchical IL-GNN-RL structure mirrors human cognition for behavior simulation. • GNN learns preference-based rewards from physical, environmental, and social features. • Modular, data-driven foundation for rapid what-if built environment analysis.

  PublisherDOI

• Human–Machine Collaboration with Reinforcement Learning for Explaining Manual Layup Strategies in Composite Materials Handling

  2026-01-28

  article

  Accurately capturing and explaining material handling strategies in civil engineering projects could help to increase safety, quality, timely delivery, as well as understanding the material waste caused by rework and environmental condition changes, especially when humans are highly involved in tasks with complex geometries. Workers must adapt their strategies based on the working environment’s characteristics, leading to substantial differences in strategies depending on context which drives downstream inefficiencies. A particularly challenging area is the handling of composite materials, as they require substantial manual forming which is inherently error prone. This work seeks to model and understand the composite material handling process. Specifically, this paper aims to enable a reinforcement learning (RL) agent to emulate humans’ adaptive material handling process through layup process monitoring and control strategies. The taxonomy system of edge conditions and actions in the paper contributes to the description of the module configuration and controls the RL model with a small search space. Predictions of the policies including the layup sequences and action are utilized by the students to perform a manual layup on the module. Thus, integration of machine-predicted strategies into human operations shows an example of human–machine collaboration, bridging the gap between human intuition and machine’s predictions. The mean total shear, wrinkled area, and total fold were reduced from 56.31% to 50.37%, 21.07% to 17.74%, and 23.80% to 22.66%, respectively, through human–machine collaboration. The policy can also be transferred to different module configurations in future works to enhance humans’ work to adapt to various environments.

  PublisherDOI

• Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

  ArXiv.org · 2026-01-20

  articleOpen accessSenior author

  Anticipating human activities and their durations is essential in applications such as smart-home automation, simulation-based architectural and urban design, activity-based transportation system simulation, and human-robot collaboration, where adaptive systems must respond to human activities. Existing data-driven agent-based models--from rule-based to deep learning--struggle in low-data environments, limiting their practicality. This paper investigates whether large language models, pre-trained on broad human knowledge, can fill this gap by reasoning about everyday activities from compact contextual cues. We adopt a retrieval-augmented prompting strategy that integrates four sources of context--temporal, spatial, behavioral history, and persona--and evaluate it on the CASAS Aruba smart-home dataset. The evaluation spans two complementary tasks: next-activity prediction with duration estimation, and multi-step daily sequence generation, each tested with various numbers of few-shot examples provided in the prompt. Analyzing few-shot effects reveals how much contextual supervision is sufficient to balance data efficiency and predictive accuracy, particularly in low-data environments. Results show that large language models exhibit strong inherent temporal understanding of human behavior: even in zero-shot settings, they produce coherent daily activity predictions, while adding one or two demonstrations further refines duration calibration and categorical accuracy. Beyond a few examples, performance saturates, indicating diminishing returns. Sequence-level evaluation confirms consistent temporal alignment across few-shot conditions. These findings suggest that pre-trained language models can serve as promising temporal reasoners, capturing both recurring routines and context-dependent behavioral variations, thereby strengthening the behavioral modules of agent-based models.

  PublisherOA PDF

• “Interaction Twin in the middle”: a distributed digital twin architecture to model team interactions and dynamics for deep space missions

  Frontiers in Aerospace Engineering · 2026-03-27

  articleOpen access

  NASA’s Moon to Mars campaign emphasizes the need for crews and habitat systems to operate with increasing autonomy as communication delays with Earth grow beyond 5 minutes. The digital twin framework has emerged as a promising solution to monitor, diagnose, predict, and optimize space systems, but prior aerospace applications have largely centered on system autonomy rather than crew autonomy. As a result, current approaches under-represent the interaction dynamics needed by mission control to continuously evolve procedure and accomplish mission objectives. This work introduces an Interaction Digital Twin (IDT) framework that twins the interactions between humans and systems rather than focusing only on individual entities. Built on a distributed digital twin architecture with bidirectional information flow, the framework integrates three complementary types of twins: Digital Twins for habitat systems, Human Digital Twins (HDTs) for individual crew members, and Interaction Digital Twins that capture emergent phenomena such as team cohesion, trust calibration, coordination, and adaptive autonomy. Twinning the interactions moves aspects of command and control on-board, giving crew mission-control-like capabilities even during periods of communication delay. We apply the framework to an Artemis Phase II mission scenario, demonstrating how interaction-level twinning extends system-level modeling to support cognitive workload management, information sharing, and human–autonomy teaming. By elevating interactions to first-class, inference-capable elements within the digital twin architecture, this framework bridges the gap between technical system models and the human teaming constructs essential for self-sufficient deep space exploration.

  PublisherOA PDFDOI

• Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

  2026-03-09

  articleSenior author
  PublisherDOI

• Expanding the Generative Power of Large Language Models for Design Through Formal Design Grammars and Languages

  Journal of Computing and Information Science in Engineering · 2025-10-09 · 3 citations

  articleOpen accessSenior author

  Abstract Research in design grammars has been underway for over 50 years and has demonstrated great generative power for a wide range of design and engineering domains. A key limitation, though, is the lack of support for designers to develop and computationally implement formal design grammars. We explore the potential of large language models (LLMs) to act as a collaborative grammar development partner that works with human designers and provides guidance during grammar development, as well as serving as a grammar interpreter that converts natural language descriptions of design grammars into executable python code. Methods for both interpreting previously known design grammars as well as interactively and collaboratively developing a new design grammar that is not known a priori are proposed. Three case studies, namely a truss design grammar, a half-hexagon shape grammar, and a technical process grammar, are investigated, covering string, shape, and graph grammars to explore the advantages and limitations of combining design grammars and LLMs. Finally, we position formal design grammars to be a key element for the future to expand the generative power of LLMs and enable them to become more repeatable, precise, and explainable for generative design tasks.

  PublisherOA PDFDOI

Recent grants

• Investigating the Effectiveness of Machine Learning Paradigms for Supporting Engineering Designers in Rapidly Evolving Digital Manufacturing

  NSF · $425k · 2018–2022

Frequent coauthors

• Jonathan Cagan

  Carnegie Mellon University

  72 shared

• Jessica Menold

  Park University

  70 shared

• Catherine Berdanier

  Purdue University System

  40 shared

• Kathryn Jablokow

  Pennsylvania State University

  37 shared

• Kenneth Kotovsky

  Carnegie Mellon University

  31 shared

• Nicolás F. Soria Zurita

  Pennsylvania State University

  26 shared

• Samuel Lapp

  University of Illinois Urbana-Champaign

  26 shared

• Nicholas A. Meisel

  Pennsylvania State University

  24 shared

Education

• Ph.D. Mechanical Engineering, Mechanical Engineering

  Carnegie Mellon University

  2016

• M.S. Mechanical Engineering, Mechanical Engineering

  Carnegie Mellon University

  2014

• B.S. Mechanical Engineering, Mechanical Engineering

  California State University, Fresno

  2012

• B.S. Civil Engineering, Civil Engineering

  California State University, Fresno

  2012

Awards & honors

• National Science Foundation Graduate Research Fellow