Brian Zhou Liu

New York, NY · brianliu0317@gmail.com · LinkedIn · GitHub

Download PDF ↗

Education

University of California, San Diego

B.S. in Data Science, AI & ML Specialization

La Jolla, CA

GPA: 3.95/4.00 · Expected Jun 2028

Technical skills

Languages
Python, Java, C++, SQL, TypeScript/JavaScript, Rust, Bash
Frameworks & libraries
PyTorch, TensorFlow, scikit-learn, Pandas, NumPy, React, FastAPI
Data & infrastructure
PostgreSQL, MongoDB, ChromaDB, Git, Docker, GCP, Cloud Run, Cloud SQL

Experience

Jan 2026 - Present

La Jolla, CA

Q-Lab, UC San Diego

Research Intern under Prof. Lianhui Qin

Building and evaluating AI systems for scientific simulation and single-cell forecasting.

  • Co-developed SIGA, a Claude Code adapter that configures scientific simulators; built its ChromaDB retrieval layer, MCP XML validator, and plugin framework, and co-authored the accompanying preprint.
  • Led a 30-task OpenFOAM transfer study; SIGA's best configuration scored 0.870 accuracy with 30/30 complete cases.
  • On harder held-out GEOS tasks, SIGA raised accuracy by 9.6% from 0.720 to 0.789 and reduced run-to-run standard deviation by about 16x.
  • Developed a direct encoder-decoder Transformer to forecast four future single-cell states from 100,000 trajectories; achieved R² = 0.536 across 3,201 output genes and R² = 0.857 on the top 50 dynamic genes.

Oct 2025 - Feb 2026

Remote

Asakana (YC F26)

Product Development Intern

Automated supplier data entry and built an AI-assisted ordering system.

  • Deployed a first-generation OCR/ETL pipeline with Gemini Flash-Lite, automating previously manual supplier entry for 10,000+ products from PDF and Excel sheets into MongoDB.
  • Developed a Dialogflow CX ordering agent backed by Cloud SQL and REST APIs, with Twilio SMS notifications and automated pricing-rule enforcement.

Sep 2024 - Dec 2025

La Jolla, CA

Rare AI Lab, UC San Diego

Research Intern under Prof. Aobo Li

Developed surrogate models that made detector-design optimization cheaper to run.

  • Implemented a multi-fidelity surrogate optimizer combining conditional neural processes and Gaussian processes, reducing detector-simulation cost by 90% for experimental design exploration.
  • Co-first-authored "Efficient Optimization of COHERENT Detector Design Parameters with RESuM," accepted to the NeurIPS 2025 ML4PS Workshop.

Mar 2023 - Nov 2023

Atlanta, GA

MAIX Lab, Emory University

Research Intern and Regeneron STS Top 300 Scholar under Prof. Ran Xiao

Built deep-learning models for cardiac screening from ECG data.

  • Constructed an anatomically informed feature tensor for cardiac screening using XResNet, achieving 93.7% AUC.
  • Boosted model sensitivity to 85.5% with 1D convolutional layers that capture cross-lead patterns.