Research

Publications & Focus Areas

My research tackles AI reliability where it matters most: critical infrastructure. From multi-agent systems that optimize power grids to methods that make LLMs more coherent and less biased—I study how to make AI that's trustworthy at scale.

Current Focus

Multi-Agent AI for Power Grids

Fellowship

VoltAge Doctoral Fellowship

Advisor

Prof. Jun Yan

Google Scholar ORCID: 0009-0005-8228-8785

Current Research

PhD focus on multi-agent AI systems for critical infrastructure.

Multi-Agent AI Systems for Power Grid Optimization

Investigating coordination mechanisms for distributed AI agents in power grid management and climate modeling. Developing novel optimization algorithms for multi-agent decision-making under uncertainty.

This work addresses the challenge of deploying reliable AI systems in critical energy infrastructure where failures have significant real-world consequences. Key research questions include agent coordination protocols, robustness under uncertainty, and scalable deployment architectures.

Multi-Agent Systems Power Grid Optimization Climate Modeling

Publications

Peer-reviewed research and software releases.

IEEE CAI 2026

SWin: A Sliding Window Summarization Approach for Coherent LLM-driven Dialogue Systems

Shreya Savant

IEEE Conference on Artificial Intelligence, Granada, Spain • May 2026

Novel memory mechanism for maintaining coherence in LLM-driven dialogue systems through sliding window summarization. Addresses the challenge of context window limitations in long-form conversations.

LLM Memory Dialogue Systems Summarization

EACL 2026

Mitigating Causal Bias in LLMs via Potential Outcomes Framework and Actual Causality Theory

Collaborative Research

European Chapter of the Association for Computational Linguistics, Rabat, Morocco • March 2026

Applies potential outcomes framework and actual causality theory to identify and mitigate causal biases in large language model outputs.

Causal Inference Bias Mitigation LLM Fairness

Published

SpokenWeb Search Engine: An Open-Source Digital Archive Search Platform

Shreya Savant et al.

Software Release, Zenodo 2024

Full-stack search platform for academic spoken word recordings, deployed at 6+ universities indexing 50,000+ cultural heritage audio recordings. Presented at Blacklight Summit 2025.

Digital Humanities Search Systems Open Source

DOI: 10.5281/zenodo.17705608

Research Threads

Long-horizon questions guiding my work.

Multi-Agent Coordination

How can distributed AI agents effectively coordinate in complex, uncertain environments like power grid management? Research explores game-theoretic approaches and emergent coordination behaviors.

PhD Focus • VoltAge Fellowship

Dialogue Memory & Coherence

How can LLMs maintain coherent, contextually-appropriate conversations over extended interactions? Investigating memory mechanisms beyond traditional context windows.

Master's Thesis • MITACS Scholar

Causal Bias Mitigation

Can causal reasoning frameworks help identify and mitigate biases in LLM outputs? Applying potential outcomes and actual causality theory to fairness in language models.

EACL 2026 Publication

Reliable AI Deployment

What architectures and practices enable reliable AI deployment in production systems with real-world constraints? Experience-driven insights from research infrastructure.

Applied Research Thread

Research Systems

Experimental platforms built to validate research hypotheses.

SpokenWeb Search Engine

Digital archive search infrastructure deployed across Canadian universities. Modernized with Docker, CI/CD, and 61% processing efficiency improvements.

Search Infrastructure Humanities AI

Publication

SWin Memory Module

Sliding window summarization approach for coherent LLM-based dialogue systems. Core contribution from master's thesis research.

LLM Memory Summarization Evaluation

Technical Details

Production RAG Pipeline

RAG systems with structured outputs, real-time ETL, and BERT-based embedding optimization. 21% retrieval accuracy improvement.

RAG ETL LLM Ops

Case Study

Education

Research training in multi-agent AI systems, dialogue memory, and LLM reliability.

PhD, Information & Systems Engineering

Concordia University, Montreal • Sep 2025–Present

Research Focus: Multi-agent AI systems for power grid optimization and climate change mitigation. Developing coordination mechanisms and optimization algorithms for distributed agents under uncertainty.

Advisor: Prof. Jun Yan VoltAge Fellowship

MASc, Quality Systems Engineering

Concordia University, Montreal • Jan 2023–Aug 2024

Thesis: Memory Mechanisms for Coherent LLM-Based Dialogue Systems. Developed novel sliding window summarization approach (SWin) for maintaining dialogue coherence in long-form conversations.

Advisors: Prof. Jun Yan, Prof. Chun Wang MITACS Accelerate Scholar

B.Tech, Computer Science & Engineering

D. Y. Patil College of Engineering, India • Aug 2018–Jun 2022

Leadership roles in technical communities. Advanced certifications in Data Analytics (IIT Kharagpur) and Applied Machine Learning (BITS Pilani).

Linux Club Coordinator Google DSC Lead

Collaborate

Open to research collaborations in multi-agent systems, LLM reliability, and energy-aware AI.