B.S. Joint Computer Science & Mathematics · Harvey Mudd College
I am currently a machine learning engineering intern at Alkera AI (YC S26), where I train SQL data agents with execution-verified reinforcement learning. I also do research with Bilin Zhuang at Harvey Mudd on unsupervised learning for molecular systems and the mean-field theory of deep networks, and with Miquel Duran-Frigola at Ersilia on tabular foundation models for low-data molecular prediction.
Research interests: LLM post-training (RL) · Tool-use agents · AI for science · Mechanistic interpretability
GMM classification in (q, LSI, sₖ, ζ) order-parameter space · likelihood-ratio abstention · TIP4P/2005, TIP5P, SWM4-NDP MD · state-resolved S(k) [2]
Deep tanh networks · mean-field variance & correlation recursions · fixed point c* · order-to-chaos boundary χ₁ = 1 · width-1,000 simulations
A null-step control, with model weights held fixed, that isolates dictionary-retraining noise in sparse-autoencoder feature-drift metrics across 112 TopK and JumpReLU SAEs on SFT and PPO checkpoints. With Cole Blodin (AWS).
A multi-agent financial research platform with real-time Server-Sent Events streams that expose agent reasoning and per-turn source-citation tracking that links research to its evidence.
An autonomous agent that combines NMR, IR, and mass-spectrometry models into ranked molecular structures with traceable evidence.
Y. Yao, D. Sanghi, A. D. Chokshi, et al. Is Water a Mixture: Unsupervised Likelihood-Ratio Classification Reveals Two Water States That Merge Upon Increasing Temperature. Under review, PNAS, 2026.
Y. Yao, C. Blodin, et al. A Null-Step Control for Sparse-Autoencoder Feature Drift Across Fine-Tuning Checkpoints. Manuscript, 2026.
GPA: 3.90
Relevant coursework: Statistical Mechanics, Probability, Statistics, Differential Equations, Graph Theory
Score: 44 / 45
Cross-bridged tetraazamacrocycle (Me₂B1#N₄) synthesis · Ni, Co, Cu complexes · Fenton-type dye-bleaching catalysis · NMR, UV-Vis, MS
