About

Naren Sivakumar

I am Naren Sivakumar, a PhD student at the University of Maryland, Baltimore County. My research focuses on robust multimodal retrieval, context-aware robustness evaluations and LLM robustness evaluations. I have built multimodal (audio + video, video only, audio only) systems to make decisions based on object and event detection, in both real-world and synthetic environments. For my master's degree, I worked on developing context-aware decision-making systems powered by a rich case history to be deployed in high-stakes negotiations. Language models, acting as agents, retrieve historical context and use it to simulate the governance of a country and formulate policies on the fly based on their counterparts' actions. As part of my applied work, I have also created several award-winning working prototypes as a part of hackathons for assistive technologies, such as voice-powered video game players and touchless phone navigation. My research is complemented by my industry experience at IBM, where I developed an AI-powered calorie-tracking application.

Projects

AutoMate

LLM-powered mobile interface navigator. Received a special mention at HoyaHax 2025.

Valkyrie

Voice-activated keypress assistant using GPT-4o. Won "Best First Hack" at HackUMBC 2024.

Truthful LLMs

Fact-checking benchmark creation and evaluation for Llama 3.2 and Gemma2.

Data Story Metrics

Developed story coherence benchmarks for LLMs in collaboration with University of Alabama.

LLM Terrarium

Multi-agent LLM simulation for emergent behavior analysis and cooperation modeling.

Prime Counter

Distributed system in Go using gRPC to count primes in large datasets.

What's New

Research & Publications

Contact