Publications
Peer-reviewed papers, thesis, and technical blog posts.
Papers & Thesis
Honest Lying: Understanding Memory Confabulation in Reflexive Agents
Investigates the phenomenon of memory confabulation in reflexive AI agents, analyzing how and why agents produce plausible but incorrect recollections, with implications for trust and reliability in long-term conversational systems.
WildfireVLM: AI-powered Analysis for Early Wildfire Detection and Risk Assessment Using Satellite Imagery
Presents WildfireVLM, an AI-driven framework for early wildfire detection and risk assessment using satellite imagery. Demonstrated at the 3rd Wildfire Digital Twin Semi-Annual Review at UMBC in front of scientists from NASA.
SOTA Results in Agentic Memory on LongMemEval
Achieved state-of-the-art performance on the LongMemEval benchmark with 86% accuracy, surpassing the previous best by 15%, using a long-term memory system for conversational agents.
SBI-RAG: Enhancing Math Word Problem Solving for Students through Schema-Based Instruction and Retrieval Augmented Generation
Proposes SBI-RAG, a framework combining schema-based instruction with retrieval augmented generation to improve mathematical word problem solving capabilities in students using LLMs.
Dynamic Edge Weighting in Relational Graph Convolutional Networks: Enhancing Sample Efficiency via Graph Attention in Reinforcement Learning
MS thesis presenting a novel approach to enhancing model-free RL algorithms by integrating graph attention networks into an R-GCN framework, achieving a 20% improvement in sample efficiency on Boxworld and Minigrid LavaGap benchmarks.
ReProHRL: Towards Multi-Goal Navigation in the Real World using Hierarchical Agents
Presents ReProHRL, a hierarchical reinforcement learning framework for multi-goal navigation, transferring from simulation to real-world robotic environments with an 85% transfer success rate.
Toward Real-World Implementation of Deep Reinforcement Learning for Vision-Based Autonomous Drone Navigation with Mission
Explores the practical deployment of deep RL for autonomous drone navigation in real-world environments using vision-based inputs and YOLO object detection, bridging the Sim2Real gap.
Peer-Reviewing Experience