Rishie Raj

I am a graduate researcher, currently working on audio-visual LLMs and multi-modal agentic systems. I graduated with a master's degree in Robotics from the University of Maryland, College Park, where most of my coarsework was in Computer Vision (CMSC472) and Multi-modal Learning (CMSC848K, CMSC848M).

Email  /  CV  /  Scholar  /  LinkedIn  /  Github

Research

My research interests primarily include multi-modal learning, audio-visual learning, and agentic systems. Below are some of my recent works.

The first MCQ-type benchmark designed to evaluate SOTA Audio-Visual Large Language Models and Omni-Modal Language Models on fine-grained cognitive tasks.

A cost-efficient neurosymbolic agent with fast-slow planning to address challenging multi-turn image editing tasks.

Other Projects

Here are some other projects that I've worked on as part of my coursework or personal interest.

A novel LLM-driven agentic framework that handles indirect, multimodal instructions without dedicated agent retraining.

A novel approach for detecting AI-generated images based on cross-perplexity and perplexity computations using autoregressive image generation models.

A small convolutional neural network model designed to predict the parameters of differentiable image processing functions with the aim to defog the input images.

An implementation of Efficient Neural Network (E-Net), finetuned for real-time semantic segmentation, which is specifically designed for tasks requiring low latency operation.

A design and implementation of a swarm multi-robot system for warehouse applications, aimed at improving safety and efficiency in material handling operations.

Design and source code from Jon Barron's website