Research

Making embeddings easier to trust.

Lab work, personal questions, and method notes around distortion, long-task memory, and reliable model behavior.

Latent Structure Lab

Fixing distortion in low-dimensional visualization

My main research continues the published work on detecting distortion in nonlinear data embeddings. Our current goal is to reduce that distortion with algorithms, so low-dimensional visualizations better preserve the high-dimensional structure they represent.

Personal Research Questions I’m Interested In

01

Multi-source hallucination

How can we reduce hallucination when a model must reconcile multiple input sources?

02

Long-task memory

How should long-short memory systems be designed when tasks are long but context windows are limited?

03

Embedding reliability

How can we quantify embedding reliability across random seeds and data shifts?

Things I’m Learning

LLM recommendation and voting

I have researched recommendation systems with large language models, including LLM + vLLM majority voting to improve answer accuracy.

Past Publications & Posters