Michael Jiao

I'm an undergrad at Harvard studying physics, computer science, and statistics.

I'm currently pursuing a joint Bachelor's degree in Physics and CS with a concurrent Master's in Statistics. Beyond my studies, I am broadly interested in philosophy, reasoning under uncertainty, and economics.

Currently an Applied AI intern at Triangle Analytics, a trader with the Harvard Quantitative Traders club, and an AI/ML research intern at CERN. Before college, I spent my time on math, physics, and astrophysics olympiads.

Outside of work, you can catch me playing soccer, poker, roaming around different restaurants, reading manga, and wasting my life on League of Legends (plat ADC main).

More on the projectsI'm building, researchI'm involved in, courses I've taken, and essays I keep returning to.

Currently

Harvard College
AB '29 · Physics + CS, AM Statistics
2025 — 2029 · Cambridge
Triangle Analytics
Applied AI intern
Oct 2025 — present · Arlington, VA
Harvard Quantitative Traders
Trader
2025 — present · Cambridge
CERN
AI/ML research intern
2026 — present · Geneva

Selected projects

all projects

  • PokerBot

    2026

    Real-time GTO poker engine. Scrapes live game state, computes optimal actions in sub-second latency via Monte Carlo equity simulation, and tracks opponent tendencies (VPIP, PFR, 3-bet%) across sessions through an async event-driven pipeline.

    Python · Playwright · SQLite · asyncio

  • Downstream

    2026

    PFAS bioaccumulation pipeline linking EPA discharge permits to personalized human health risk across 3,274 river segments. Gradient boosting predicts water concentrations (80.6% within 3×); a physics-informed neural network models tissue uptake via the Gobas ODE (R² 0.94). Surfaced through an interactive Mapbox UI with species- and consumption-aware hazard quotients. Built in 16 hours at YHacks 2026.

    React · FastAPI · PyTorch · scikit-learn · Mapbox

  • ML4Sci / DeepLense (GSoC '26)

    2026

    Evaluation tasks for the ML4Sci DeepLense Google Summer of Code project on domain-adaptive lens finding in wide-field surveys. Modified ResNet-18 classifies dark-matter substructure across three simulated classes (macro-AUC 0.9896); EfficientNet-B0 distinguishes lensed from non-lensed galaxies under heavy class imbalance on g/r/i cutouts (AUC 0.9808).

    PyTorch · ResNet · EfficientNet · NumPy

  • GANaxy

    2025

    Anomaly-detection pipeline using DCGAN, f-AnoGAN, and SimCLR to classify ~300K Galaxy Zoo 2 images at 94.7% accuracy. PCA/UMAP visualization surfaced ~8,500 anomalies — mergers, starbursts, artifacts.

    PyTorch · NumPy · pandas · scikit-learn

Honors

  • Honorable Mention, Team USA
    International Mathematical Modeling Challenge
    2024
    Best paper out of 200+ U.S. teams; top 40 of 2,500+ globally.
  • Silver Medalist, Finalist
    Bridgewater Forecasting Competition
    2026
    Top 20 of 1,300+ undergrads; 36 of 2,600+ globally.
  • AIME Qualifier
    Mathematical Association of America
    2021 — 2025
    Top 5% of 300,000+ in competition mathematics.
  • Bronze Medal
    USA Physics Olympiad
    2024 — 2025
    Top 175 nationally.
  • IOAA TST Invitee, Team Canada
    Canadian Astrophysics Olympiad
    2024
    Top 10 nationally; one of 15 at training camp.

What I’ve been reading

all reads
  • The Bitter Lesson — Rich Suttonessay · 2019
  • Flash Boys — Michael Lewisbook · 2014
  • The Market for 'Lemons' — George Akerlofpaper · 1970
  • Reflections on Trusting Trust — Ken Thompsonpaper · 1984