Tianhao Cao
  • Home
  • Projects
  • Resume
  • About
Categories
All (8)
Causal Inference (1)
Classification (2)
Corpus Linguistics (1)
Deep Learning (1)
Econometrics (2)
Ensemble Methods (3)
Explainability (1)
Information Retrieval (1)
LLM (2)
Machine Learning (3)
Multimodal (1)
NLP (5)
Privacy & Security (2)
Python (6)
R (2)
RAG (1)
Regression (2)
Security (1)
VLM (1)

M.S. Data Science · Computational Linguistics

Turning statistics into
working software.

I’m a master’s student at the University of British Columbia, working where statistics meets computer science. My interests are data science, machine learning, and natural language processing — and I like building tools that answer a real question, whether that means finding the pattern under a phenomenon or writing the model that predicts it.

View my projects See resume

Tianhao Cao
Vancouver, BC · University of British Columbia

Currently an applied machine learning engineer intern on the Ednoda × UBC MDS-CL capstone, building an end-to-end curriculum pipeline for an ESL platform: grammar analysis, semantic retrieval, and automated teaching-material generation.

01 Selected work

Projects

Six pieces from the master’s programme in NLP, machine learning, and privacy — and two from undergraduate economics.

Master’s

FraudCorpus: Multilingual Social Engineering Detection
Python
NLP
Machine Learning
Security
Corpus Linguistics

An end-to-end NLP pipeline for detecting phishing, spam, and social engineering attacks across English and Chinese text

Tianhao Cao
Apr 28, 2026

Membership Inference Attack on Fine-tuned VLMs
Python
VLM
NLP
Privacy & Security
Multimodal
Ensemble Methods

Multimodal attack pipeline combining CLIP embeddings, loss-ratio signals, and neighborhood perturbation to detect training data membership in vision-language models

Tianhao Cao
Apr 18, 2026

Betty Talker — Recipe RAG Pipeline
Python
NLP
RAG
LLM
Information Retrieval

A full RAG pipeline that answers recipe questions and scales ingredients, built on Betty Crocker’s Bisquick Cook Book using LLMs and semantic search.

Tianhao Cao
Mar 25, 2026

Membership Inference Attack on Fine-tuned LLMs
Python
LLM
NLP
Privacy & Security
Ensemble Methods

Dual-model feature fusion to detect training data membership in fine-tuned language models

Tianhao Cao
Mar 18, 2026

Sentiment Analysis on Yelp Reviews
Python
NLP
Deep Learning
Classification

Comparative analysis of CBOW and LSTM models for sentiment classification on Yelp reviews

Tianhao Cao
Feb 1, 2026

Credit Card Default Analysis
Python
Machine Learning
Classification
Ensemble Methods
Explainability

Using XGBoosting to predict credit card defaults

Tianhao Cao
Dec 11, 2025
No matching items

Undergraduate

Weather Insurance Purchasing Prediction
R
Machine Learning
Regression
Econometrics

Predicting farmers’ insurance buying behavior using Lasso and Random Forest

Tianhao Cao
Jun 5, 2022

Covid-19 DiD Analysis
R
Causal Inference
Regression
Econometrics

Identifying the influences of Early-Policies to the COVID-19 transmission and Deaths

Tianhao Cao
Mar 18, 2022
No matching items

02 Record

Resume

Two degrees, one capstone, and a toolkit built around language models and the data pipelines that feed them.

Download PDF

Education

Two cities, one thread

University of British Columbia

Vancouver, BC · Aug 2025 – Present

M.S. in Data Science, Computational Linguistics. Supervised and unsupervised learning, natural language processing, machine learning, statistics.

University of California, Santa Cruz

Santa Cruz, CA · Sep 2019 – Jun 2022

B.A. in Business Management Economics. Advanced quantitative analysis, machine learning economics, statistics, security markets.

Technical Skills

The toolkit
Category Skills
Languages Python, R, SQL, MongoDB, Bash
ML & AI PyTorch, Scikit-learn, XGBoost, Hugging Face Transformers, BERT, RoBERTa, Flan-T5, sentence-transformers, LLM fine-tuning, RAG, multi-agent LLM, vector embeddings, prompt engineering, SHAP
NLP Word embeddings, n-gram models, parsing algorithms, membership inference, NLTK, spaCy, CEFR
Data Eng. Pandas, NumPy, Apache Parquet, ETL pipelines, feature engineering, data visualization
MLOps Docker, AWS, Git/GitHub, CI/CD, Weights & Biases, LangChain / LlamaIndex
Tools VS Code, Jupyter, Quarto, LaTeX

Experience

Where the theory met a deadline

Applied Machine Learning Engineer Intern

Ednoda × UBC MDS-CL Capstone · Jan 2026 – Jun 2026

Built an end-to-end AI-powered curriculum content pipeline for an ESL edtech platform, spanning grammar analysis, semantic retrieval, and automated teaching-material generation.

  • Designed 29 spaCy DependencyMatcher grammar templates to classify ESL sentences by syntactic structure — 65.8% coverage on production data, 54.2% on the CEFR-SP benchmark (A1–C2 consistent).
  • Built an offline embedding pipeline with all-MiniLM-L6-v2, encoding 6,477 grammar-tagged sentences into a searchable semantic index.
  • Fine-tuned a BERT model for CEFR difficulty classification (A1–C2), reaching 92% (±1) accuracy.
  • Engineered a three-layer retrieval system: grammar-structure filtering, CEFR difficulty matching, and cosine-similarity ranking.
  • Built a three-layer generation system producing new practice sentences through slot detection, masked-token candidates (RoBERTa for single tokens, T5 / Flan-T5 for spans), and cosine ranking.
  • Engineered a PDF parsing pipeline that turns textbook question sections into machine-readable datasets, plus a RAG system that cuts redundancy across generated content.
  • Designed a multi-agent LLM workflow that ingests a textbook PDF and produces a complete teaching package: lesson notebook, practice questions, graded quiz.

Career Mentorship Program — Mentee

With Jacob Oh · Jan 2026 – Present

Selected for a professional mentorship program under an industry veteran. Software engineering best practices, career development in tech, and translating academic work into industry standards.

Capstone stack — Python · spaCy · sentence-transformers · BERT · RoBERTa · Flan-T5 · LangChain / LlamaIndex · RAG · multi-agent LLM · pandas · Apache Parquet · CEFR

03 Method & interests

A fascination with patterns

My Journey

Statistics asked, CS answered

My academic work is driven by patterns — hidden in large datasets, embedded in human language, or structured inside musical compositions. Statistics taught me to ask whether a pattern is real. Computer science taught me to build something that acts on the answer.

Beyond the Code

When I’m not writing Python or analysing data, I’m usually working through music theory — how chords and scales function mathematically. It complements the coding more than it looks: both ask you to recognise structure and solve problems creatively inside a set of rules.

Technical Philosophy

Not a black-box user

I don’t just run code. I want to understand the mathematics behind the algorithm.

  • In data science — a statistics and CS background means understanding model behaviour, not treating it as a black box.
  • In economics — micro and macro theory combine with data science to frame questions a model alone would miss.

Interests

  • Programming — Python (Pandas, NumPy, ScikitLearn, PyTorch), R, SQL
  • Data science — machine learning, data analysis, NLP (word embeddings, parsing)
  • Other — music theory, mathematics
Home Projects Resume About