About Me

I’m a third-year PhD student in the Data Systems Group (DSG) at MIT where I am advised by Tim Kraska. I am a recipient of the 2025 JPMC AI Research PhD Fellowship.

My research is focused on building an AI query engine for Deep Research and unstructured analytics tasks more generally. Interest in Deep Research has grown considerably, with OpenAI, Anthropic, xAI, Perplexity, and Google rolling out offerings to their combined millions of users. The allure of Deep Research is the promise of its simplicity: upload your data, ask a question in natural language, and get an answer. However, Deep Research systems—which augment LLMs with the ability to reason, use tools, and execute code—can fail to deliver on this promise for even relatively simple queries. Furthermore, the cost of their execution can be orders of magnitude higher than standard invocation of an LLM. The goal of my research is to improve the quality, cost, and latency of these systems.

In my first year, I was co-first author and a lead developer for our work on Palimpzest: a Python programming framework and query engine for writing and optimizing declarative AI programs. Palimpzest extends the DataFrame execution model to include LLM-powered maps, filters, and joins whose semantics are specified in natural language. We presented our vision paper at CIDR’25 and showed that Palimpzest could execute AI programs 3.3x faster and 2.9x cheaper than a baseline method, while also achieving superior output quality.

Building on our vision paper, my second paper on Abacus (accepted to VLDB’26) applied cost-based optimization to this setting, enabling Palimpzest to optimize execution plans for cost, latency, and/or quality. We demonstrated that, on average, systems optimized by Abacus achieved 6.7%-39.4% better quality and were 10.8x cheaper and 3.4x faster than the next best system.

Most recently, I submitted a vision paper (accepted to CIDR’26) which extends Palimpzest to support and optimize Deep Research queries. Our preliminary results demonstrate that our query engine can improve the cost and quality of execution plans generated by Deep Research agents, with cost and runtime savings of up to 76.8% and 72.7% respectively.

Before coming to MIT, I worked at Cambridge Mobile Telematics (CMT) while also pursuing a Master’s degree at Stanford under the wonderful guidance of Daniel Kang and Matei Zaharia.