Vijay Murari Tiyyala

Hi, I’m a recent Master’s graduate in Computer Science from Johns Hopkins University, specializing in Machine Learning and NLP. Interested in my work? Check out my resume.

In my research at the Center for Language and Speech Processing, I worked under Prof. Mark Dredze to improve chatbot interactions in healthcare. Before this, I developed a multilingual medical concept dictionary with Dr. David Yarowsky. I also have experience as a Data Analyst at Deloitte before pursuing my Master’s.

Having completed my Master’s in December 2023, my academic journey involved deep dives into Deep Learning, Advanced NLP, and ML System Design. My undergraduate studies were in Computer Science at Velagapudi Ramakrishna Siddhartha Engineering College.

Outside of research, my interests include classical music, moderate gaming, exploring Dravidian languages, and watching sci-fi thrillers like the Terminator series.

This site is where I share insights from my academic journey and thoughts on the latest in NLP.

Projects

Empathy-Enhanced LLMs: Refining AI Responses with Fine-Grained Human Feedback

In this project, we are fine-tuning models for controlled text generation, focusing on empathy, quality, factuality, and comprehensiveness. Our approach is closely aligned with the principles outlined in the paper “Fine-Grained Human Feedback Gives Better Rewards for Language Model Training.” This involves training an ensemble of reward models, each tailored to a specific style of text response. By leveraging fine-grained human feedback, we can refine chatbot interactions in healthcare, ensuring responses are not only accurate and relevant but also empathetic and comprehensive.

SAMOYEDS is a simulation tool designed to anticipate and shape the public response to vaccination strategies and other public health policies. Developed in the context of the COVID-19 pandemic, this project utilizes Mistral 7B, a generative AI model, to simulate human-like behavior patterns within a social network mirroring the U.S. demographics. The project’s uniqueness lies in its ability to dynamically model human responses, providing invaluable insights for policy-making in public health. By successfully navigating the complexities of COVID-19 policy, SAMOYEDS demonstrates its potential as a versatile tool for future challenges in public health governance.

Medical Terminology Translation and Multilingual Matrix Construction

This initiative involves creating a massive multilingual matrix for medical terms, focusing on the translation and analysis of compound words. Inspired by the paper “Massively Translingual Compound Analysis and Translation Discovery,” the project aims to understand and translate complex medical compounds across various languages. We analyze compounds in over 300 languages, identifying patterns in word formation and semantics. This work is crucial for enhancing machine translation, especially in low-resource languages, and provides a robust foundation for further research in compound word processing and translation.

Improving Code Editing through Self-Instruct

I worked on diversifying methods of code editing by implementing novel data augmentation strategies, enhancing the efficiency of code editing processes, and fine-tuning code LLMs.

Please feel free to reach out to me for potential collaborations.