Andy Huang
CVGitHubLinkedIn

Skills

Languages

PythonJavaJavaScriptSQLHTML/CSSRC

Frameworks

ReactNext.jsFastAPITailwind CSSOpenAPINode.js

Libraries & Tools

PandasNumPyMatplotlibSeabornOpenCVGitPrismMicrosoft FabricSQL Server

Technologies

Machine LearningData AnalysisData Visualization

Andy Huang

About Me

I'm a graduate from the University of California San Diego, San Diego, CA with a B.S. in Data Science.

I work at the intersection of data engineering, machine learning, and software development, with interests in LLM inference, brain-computer interfaces, and scalable data pipelines.

Experience

San Diego Association of Governments (SANDAG)

Current
Data Science Intern · San Diego, CA
June 2025 - Present

Built a cloud data warehouse in Microsoft Fabric from scratch using a Lakehouse architecture to support scalable data ingestion, transformation, and storage. Developed Dataflow Gen2 pipelines to migrate data into a structured, maintainable environment for analytics and reporting. Designed ingestion and transformation workflows improving data accessibility and reliability across datasets. Queried and validated large datasets using SQL Server, implementing reconciliation checks to ensure reporting accuracy. Automated CSV processing, email data extraction, and report generation using Python, reducing manual workload by ~10 hours/week.

Projects

LLM Speculative Decoding on TPU (DFlash)
Sep 2025 – Feb 2026 · University of California, San Diego

LLM Speculative Decoding on TPU (DFlash)

  • Ported diffusion-based speculative decoding from PyTorch/GPU to JAX/TPU within the vLLM inference framework.
  • Achieved 2.85x serving speedup (265 TPS) and up to 3.02x standalone speedup (773 TPS) on TPU v4/v5p.
  • Diagnosed and fixed sequence-length alignment bug, improving acceptance length from 2.49 to 4.75 and speedup from 1.30x to 2.85x.
  • Contributed production-ready code including model integration, proposer logic, benchmarking tools, tests, and CI.
GitHub
JAXTPUvLLMPyTorchPython
Brain-Computer Interface (BCI) Mouse via EEG
Jan 2025 – Apr 2025 · University of California, San Diego

Brain-Computer Interface (BCI) Mouse via EEG

  • Developed a BCI-powered cursor control system using EEG signals and SSVEP classification for hands-free interaction.
  • Achieved 100% directional accuracy across 25 trials through optimized signal processing and classification pipelines.
  • Improved EEG data collection efficiency by 15% through enhancements in experimental setup and data acquisition.
EEGSSVEPSignal ProcessingPythonMachine Learning
League of Legends Data Analysis & Baseline Model
Jun 2023 – Jul 2023 · University of California, San Diego

League of Legends Data Analysis & Baseline Model

  • Built a logistic regression model using gameplay statistics from 50,000+ matches to predict match outcomes.
  • Performed univariate and bivariate analysis to identify key features influencing outcomes.
  • Improved model accuracy by ~10% while analyzing bias and feature importance.
PythonPandasNumPyMatplotlibMachine Learning
Website
Download Resume