Portrait of Sepehr Ilami

Sepehr Ilami

PhD Student in Industrial Engineering • Northeastern University

Multi-agent Systems Computational Social Science Collective Behavior

Welcome! I'm a fourth-year PhD student in the department of Industrial Engineering at Northeastern University. My research lies in Multi-agent Systems, Computational Social Science, and Collective Behavior. I'm fortunate to be advised by Babak Heydari. Previously, I got my BSc in Computer Engineering at Aryamehr (Sharif) University of Technology in Iran, where I was born and raised.

CV Google Scholar GitHub Twitter LinkedIn Email

Research

My research is in Network Science and Multi-Agent Systems. I explore how intelligent entities—either humans or AI agents—coordinate and influence each other in networked environments. I work across two main areas:

Multi-agent LLM systems

Building frameworks for distributed AI agents to collaborate, share information, and make collective decisions; studying incentives, governance, and robustness in agent societies.

Human mobility & collective behavior

Modeling how people move through networks and interact in urban settings; analyzing coordination, compliance, and emergent patterns during routine and high-stakes situations.

My work sits at the intersection of computer science, social science, economics, and statistics—broadly described as network science.

Research Papers

Four observation-network topologies at a matched budget: hub, regular, expander, and modular barbell
ArXiv Preprint • 2026

Steering Emergent Behavior in Multi-Agent LLM Systems via Information Networks

Holding agents, payoffs, and the interaction network fixed, we make the information network—who observes whom—the design object for a collective of LLM agents. Observation acts as a behavioral-coupling channel whose welfare sign is set by alignment: across six repeated games, more peer information helps when individually attractive behavior serves the collective goal but amplifies free-riding and congestion when it does not, and there a modular “barbell” wiring is reliably the worst topology.

Authors: S. Ilami and B. Heydari.

arXiv link coming soon
Networked risk perception and peer effects in social systems: the case of a pandemic
ArXiv Preprint • 2026

Networked Risk Perception and Behavioral Bubbles: the case of a pandemic

Rather than a purely individual cognitive readout of objective hazards, this paper shows that risk perception is socially constructed at the community level through human mobility networks. Our analysis reveals that behavioral spillovers require both routine physical intersection and socioeconomic similarity to propagate, establishing mobility-defined communities—rather than administrative borders—as the operative scale of crisis response.

Authors: S. Ilami, M. Comola, S. Prina, and B. Heydari

Paper
Network opacity framework for The Architecture of Illusion project
ArXiv Preprint • 2026

The Architecture of Illusion: Network Opacity and Strategic Escalation

We study how strategic actors exploit hidden network structure to shape escalation dynamics. The analysis connects information shielding and perceived connectivity to when conflicts intensify or stabilize.

Authors: R. Ebrahimi, S. Ilami, B. Heydari, I. Trevino, and M. Franceschetti.

Paper
Adaptive information modulation framework for multi-agent AI governance
Journal Article • 2026

Adaptive Information Modulation: Designing Governance Mechanisms for Multi-Agent AI Systems

We design governance mechanisms that adaptively shape information exposure among AI agents. The framework aligns incentives, improves coordination, and is validated through multi-agent simulations.

Authors: Q. Chen, S. Ilami, N. Lore, and B. Heydari.

Paper Code
LLM fine-tuning workflow for strategic reasoning transfer
Journal Article • 2026

Transferring Strategic Thinking From Large to Small Language Models for Multi-Agent Systems

Fine-tuning a small model on a large model's answers and stated reasoning over canonical social dilemmas transfers game-theoretic behavior, not just surface style: behavioral alignment with the larger model rises 46% on average, holds up 18% in unseen social contexts, and survives re-parametrization of payoffs—making agentic swarms of small models viable.

Authors: N. Lore, S. Ilami, and B. Heydari.

Paper Code
Mobility network structure figure for epidemic forecasting project
ArXiv Preprint • 2025

Network Topology Matters, But Not Always: Mobility Networks in Epidemic Forecasting

We examine when mobility network structure meaningfully improves epidemic forecasts. The results show which topological features help prediction and when aggregate mobility is sufficient.

Authors: S. Ilami, Q. Cao, and B. Heydari.

Paper
AI-driven COVID-19 diagnosis workflow using medical imaging
Book Chapter • 2022

COVID-19 Diagnosis with Artificial Intelligence

We develop AI-driven diagnostic systems for rapid COVID-19 screening using CT scans and X-rays. The pipeline combines model design, rigorous validation, and explainability to support clinical reliability, while extending to resource management, remote monitoring, and contact tracing.

Authors: R. Ghavami Modegh, A. Salimi, S. Ilami, A. H. Dehqan, H. Dashti, S. Haghjooy Javanmard, H. Ghanaati, and H. R. Rabiee.

Paper

Working Papers

  1. "Communication, Belief Dynamics, and Trust Formation in Decentralized Agentic Markets"

    Studying how AI agents communicate, form beliefs, build trust, and respond to governance interventions in decentralized market-like environments.

Conference Presentations

  1. "When Does Dynamic Behavioral Information Improve Sociotechnical Modeling?"

    TalkInternational Conference on Computational Social Science (IC2S2)

    July 2026 University of Vermont

  2. "Mobility Networks Make Up for Missing Local Data in Epidemic Forecasting"

    TalkInternational Conference on Network Science (NetSci)

    June 2026 Northeastern University

  3. "Adaptive Information Modulation: Designing Governance Mechanisms for Multi-Agent AI Systems"

    Poster2026 Research Expo - Northeastern College of Engineering

    Feb 2026 Northeastern University

  4. "Integrated Design and Governance of Agentic AI Systems through Adaptive Information Modulation"

    Poster2025 Research Expo - Northeastern MIE Department

    Dec 2025 Northeastern University

  5. "Instigating Cooperation among LLM Agents Using Adaptive Information Modulation"

    TalkInternational Engineering Systems Conference (CESUN 2025)

    June 2025 George Mason University

  6. "Integrating Language Models with Agent-Based Modeling for Policy Analysis"

    Poster2024 Research Expo - Northeastern MIE Department

    Dec 2024 Northeastern University

  7. "Integrating Language Models with Agent-Based Modeling for Policy Analysis"

    PosterInternational Conference on Computational Social Science (IC2S2)

    May 2024 University of Pennsylvania

Get In Touch

I’m open to collaborations in computational social science, network science, and multi-agent systems. If you're also interested in these areas or have any questions, feel free to drop me an email or connect on LinkedIn! :-)

📧 ilami.a@northeastern.edu
📞📍 Northeastern University, Boston, MA

Google Scholar LinkedIn GitHub