I build intelligent systems—and study how to know when they actually work.

My work connects large-scale machine learning engineering for personalized recommendation with rigorous evaluation of AI agents, tools, and workflows, grounded in statistical reasoning.

At a glance · Vancouver / UBC

I’m Jiaping Liu. I also go by Olivia. I build machine learning systems and design rigorous ways to evaluate them, with a particular focus on personalized recommendation, ranking, and retrieval. My experience spans production recommender systems at JD.com, large-scale risk modeling at RBC, and open-source statistical software and computational research through my Statistics PhD at UBC. I’m especially interested in large-scale online ML systems, multimodal and agentic models, and evaluation pipelines that turn model behavior into reliable evidence for product decisions.

Applied evidence

Applied at scale.

Scroll horizontally for evidence

01E-commerce

JD.com

Machine Learning Engineer (Algorithm Development Engineer)

Personalized recommendation system

Improved click-through rate by ranking retrieved products for push and messaging recommendations, combining feature engineering, deep learning, joint pointwise and pairwise learning, and large language models; achieved a +0.31% relative overall pCTR increase and +0.78% among high-active customers.
Built SQL monitoring panels for online and offline metrics, conducted A/B experiments, explored natural-language query understanding, and addressed online-offline inconsistency and cold start for low-active users.

An independently built, non-confidential example inspired by this internship context.Modern personalized recommendation ↗

02Financial services

RBC

Risk Modeling Analyst

Credit risk modeling pipeline

Built a credit-risk modeling pipeline on billion-level user-behavior data, using survival analysis as a baseline to predict probability of default for overdraft accounts.
Resolved large-scale data-quality issues involving duplicated records, substantial missingness, and incorrect account open, reopen, and close dates.
Explored LLM-derived transaction embeddings and reimplemented survival-model fitting in PyTorch, reducing training time by more than 4× on million-row data with distributed scaling support.

03Higher education & research

UBC

PhD Candidate in Statistics

Computational statistics research

Methodology, computational optimization, open-source research software, writing, and teaching.

04Research consulting

UBC ASDa

Statistical Consultant

Statistical consulting across research domains

Collaborated with more than 10 clients across research domains, applying statistical modeling and machine learning and independently delivering data-driven analysis reports.Applied Statistics and Data Science Group (ASDa) ↗

05Government