Joy Arulraj

We design multi-modal data systems that make AI applications more efficient, expressive, and usable.

We are curating the Periodic Table of Computer System Design Principles, an effort to distill cross-cutting design principles from computer systems research into a shared vocabulary for students, researchers, and practitioners.

We developed EvaDB, an AI-relational data system for multimodal data that addresses the challenge of optimizing query execution when model inference is expensive and data is unstructured.

We are now building TokenSmith, a system for question answering over textbook-style educational content that addresses the challenge of accurate retrieval and reasoning over long documents using local language models.

I am looking for a few graduate students interested in data systems and AI. If you are interested, we should talk. Please fill out this research interest questionnaire. Georgia Tech undergraduate students are also encouraged to reach out.

I am teaching a new OMSCS course on building database systems, based on BuzzDB, an educational database system developed at Georgia Tech.