Jingxian Liao

Jingxian Liao

PhD Candidate @ University of California, Davis

Working experiences

PhD Researcher

University of California - Davis
Davis, CA, USA
Sep 2018 - Present

  • Applied AI: Led a BERT-based personalized video recommendation algorithm in Python, Uplifting user satisfaction by 14.4% compared to default YouTube list
  • Full-stack and Design: Developed scalable GPT-aided, full-stack Web systems using React, JavaScript, and D3.js (frontend), and NoSQL, Python (backend) which increased human-AI interactions and improved click rates by 2x
  • DS: Analyzed 10k comments across videos, from data cleaning, labeling to visualization, modeling, to gain insights from crowd wisdom for video recommendation
  • Experiments: Conducted ethnography, surveys, and interviews, executing lab experiments with 200+ participants, providing actionable insights to design and engineering practitioners and researchers
  • Leadership: Led multiple undergrad research projects and mentored six junior researchers across ML/AI/HCI research

Research Intern

Google Research
Mountain View, CA, USA
Jun 2022 - Sep 2022

  • Engineering: Optimized an automated system for statistical hypothesis testing and data analysis, reducing health experts’ most coding efforts, adaptable for most statistics methods, with logging, automated testing, code reviews
  • ML: Created latent representations of 500k data via geographical analysis and UMAP, to reduce hundreds of raw feature lists into explainable lifestyle dimensions for model explainability
  • Statistics: Employed quantitative regression and Bayesian techniques to guide future public health research start points

Data Scientist Intern

Intuit AI, Intuit
Mountain View, CA, USA
Jun 2020 - Sep 2021

  • Engineering: Streamlined an automatic preprocessing algorithm based on 10M bank records to identify financial health and loan risk using Spark, SQL, and Databricks, resulting in a 37% improvement in the accuracy of loan applicants’ annual revenue predictions
  • ML with Patent And Deployment: Deployed and developed self-training models for credit score assessment which boosted default risk prediction accuracy by 4% and increased data size by 26% and now employed by underwriters to refine underwriting process
  • Fintech: Innovatively defined adjusted approval rate (a KPI) to gear towards risk slope optimization, enhancing financial risk control capabilities by increasing the loan approval rate up to 8.8% with the same default rate