Bowen Yu 余博文 IPA: [y˧˥ po˧˥ wən˧˥]
Senior undergrad, MIT Physics & AI
I'm an undergraduate at MIT (Class of 2027), double-majoring in Physics (Course 8) and Artificial Intelligence (Course 6-4). Before MIT I spent a year studying physics at Peking University.
My research interests center around building intelligent systems that can efficiently learn and adapt in complex environments. I'm currently working with Prof. Yilun Du at the Embodied Minds Lab on the dynamics of multi-agent systems, supervised by Zhenting Qi. Previously, I worked with Prof. Mingda Li at the MIT Quantum Measurement Group on machine learning for condensed matter physics, materials discovery, and defect engineering, supervised by Mouyang Cheng.
Outside the lab, I enjoy long-distance running, classic films, linguistics (including learning all sorts of languages; I'm currently studying Japanese, Spanish and Cantonese!), rhythm games (Phigros, CHUNITHM), philosophy, vibe-coding projects, and Liu Cixin's novels.
Selected publications
All publications-
A Foundation Model for Non-Destructive Defect Identification from Vibrational Spectra
† Equal contribution. * Corresponding author.
Matter
A non-destructive defect identification probe to predict the chemical identity and concentration of substitutional point defects directly from vibrational spectra
-
Probing Non-Equilibrium Grain Boundary Dynamics with XPCS and Domain-Adaptive Machine Learning
† Equal contribution. * Corresponding author.
In review
Pairing X-ray photon correlation spectroscopy (XPCS) with semi-supervised domain adaptation to quantify slow, non-equilibrium grain-boundary dynamics in nanocrystalline silicon
Experience
View full experience-
Citadel Securities
Quantitative Research Analyst
Systematic alpha research on equity factor return prediction with ML models.
-
ByteDance
Large Language Model Data Researcher
Built a 500+ problem physics benchmark, identify the perception/reasoning badcases for frontier LLM within physics context, and a test-time scaling pipeline that achieved gold medal performance in IPhO 2025.
-
Intel Corporation
AI Workload Deployment Intern
Developed GPU memory-layout tutorials and CuTeDSL performance references for AI workloads.
Highlighted posts
All posts
Jul 3, 2025
Linear Layouts in Triton
Notes on linear layouts in Triton and its conversion with various traditional layout types.
May 3, 2026
Can God Be Proved to Exist? —Regarding Anselm’s Ontological Argument
Anselm’s ontological argument, presented in Proslogion II, seeks to demonstrate the existence of God solely by reflecting on the concept of God as "something than which nothing greater can be thought." I will analyze the argument, its premises, and the logical structure in this blog post.
Highlighted projects
All projects
Sep 20, 2026
Sparkie (HackMIT 2026)
An AI meeting agent that listens, speaks, and GETS WORK DONE in video calls.
Mar 17, 2026
Daily Logs
An iOS daily life logging app for tracking sleep, meals, showers, notes, photos, and lightweight personal analytics.
Apr 25, 2026
Fireroad.ai (MIT CSAIL Agentic AI Hackathon 2026)
An AI-powered course-planning prototype for MIT students, combining a Fireroad-style planner with a tool-calling academic advising chat agent.