Projects - search

Professor T. Brown recently developed a system called HeapLENS to help researchers automatically examine the memory layout of multithreaded applications. HeapLENS is specifically designed to produce compact, high-quality, curated output suitable for AI-driven analysis. While HeapLENS output can already enable AI agents to improve application memory layouts by a significant margin, the current workflow invokes HeapLENS only once and uses its output only once. A natural research direction is therefore to adapt HeapLENS to support repeated interaction with an AI agent, enabling an iterative optimization cycle in which incremental changes can be proposed, evaluated, and refined.

Tags: C/C++, Data Structures, Multithreading, Memory Management, Operating Systems, Systems, Artificial Intelligence, 2nd Year +

LLM-based agents are now used to write tests and fix bugs in real software.  They sometimes succeed but when they fail, we usually have no idea why. Every agent leaves a full step-by-step log of what it did, called a trajectory. Many of these logs are now public, but almost no one has sat down and studied them carefully. This project analyzes those logs to understand how agents actually work in generating tests, where they get stuck, and what makes some tasks harder than others. This matters because developers are starting to trust these tools with real work. If we understand how and when they fail, we can build better tools and know when their output needs a second look. Recent public benchmarks like SWT Bench and SWE Atlas, built on real open-source projects, release these trajectories openly, so the data is ready to use.

Tags: Basic Programming, Python, Artificial Intelligence, All Years

This project explores how AI can help understand bilingual doctor–patient conversations and automatically generate accurate medical documentation. It has the potential to improve healthcare accessibility and reduce documentation workload for clinicians serving multilingual populations. We have already build 280 hours speech corpus containing code-switched Kazakh-Russian medical data.  We now collecting an additional 100 hours of simulated doctor and patient conversations to improve model performance.

Tags: Basic Programming, Python, Artificial Intelligence, Machine Learning, Natural Language Processing, Data Science, All Years

Modern AI and machine learning systems are increasingly trained and deployed on distributed infrastructures consisting of multiple servers working together. While distributed computing enables larger models and faster processing, it also introduces new security challenges. Communication between nodes, shared resources, and distributed coordination mechanisms can create vulnerabilities that may not exist in single-machine systems. The goal of this project is to understand and evaluate security risks that arise when training or running AI/ML models in distributed environments. By identifying and studying these vulnerabilities, we can help build more secure and trustworthy AI systems.

Tags: Networks, Operating Systems, Artificial Intelligence, Machine Learning, Security, Systems, All Years

Healthcare data can reveal important insights that improve patient care, but analyzing it is challenging. Analysts must explore complex datasets, generate and test hypotheses, and interpret results carefully. While Generative AI can assist by creating code, visualizations, and insights, it does not always understand users' goals and can sometimes produce unreliable results. This project explores how teams of AI agents can collaborate with humans to support healthcare data analysis. We will design new interaction techniques that help people communicate their intent, understand how AI-generated results were produced, and assess whether those results are trustworthy. By making human-AI collaboration more transparent and reliable, this research aims to help healthcare professionals gain insights from data more effectively and make better-informed decisions.

Tags: Web Development, Data Analysis, Human Computer Interaction (HCI), Artificial Intelligence, All Years

AI coding agents can attempt real compiler work, but they stumble on implementing optimizations: asked to add a rewrite rule to LLVM's InstCombine pass, they often produce patches that miscompile programs, break tests, or land in the wrong place, and our benchmarking shows agents fail many such tasks. The open question is what feedback closes the gap: when the agent is handed a correctness counterexample, a profitability estimate, or a regression result, does its success rate improve, and which helps most? This project answers that on a fixed open model in a fully observable loop.

Tags: Compilers, Artificial Intelligence, Python, Command Line, C/C++, 2nd Year +, Experienced 1st Years