Kyle Whitecross

Kyle Whitecross

PhD Student in CIIR Lab @ UMass Amherst | Research Intern @ Snowflake

Hi!

I’m Kyle Whitecross, a second-year PhD Student at the Manning CICS in UMass Amherst. I work at the Center for Intelligent Information Retrieval, where I am advised by Prof. Negin Rahimi. This summer, I’m a research intern at Snowflake on the Cortex team, working on analytical search over unstructured data. Previously, I was a B.S. student at UCLA in the Samueli School of Engineering, where I studied Computer Science.

Research

I work on agentic LLMs for complex, information-intensive tasks that unfold over long reasoning trajectories. My current focus spans three interlocking directions: long-horizon reasoning, in-context retrieval, and agentic systems with persistent memory. Most recently, in RecaLLM (CoLM 2026), we identified lost-in-thought, a failure mode where chain-of-thought reasoning degrades a model’s ability to retrieve information from its own context, and addressed it by post-training LLMs to interleave reasoning with explicit, verifiable in-context retrieval. In the past, I have also worked on designing more expressive Graph Transformers with Prof. Rex Ying, compiling neural networks into tractable boolean circuits with Prof. Adnan Darwiche, and studying robustness to noisily labelled data with Prof. Baharan Mirzasoleiman. I also interned at Kumo.ai where I developed and deployed a method to automatically identify target leakage in large industry datasets.

News