Zhaoyan Dong

MLE @Moloco| Master of Engineering @UCLA | Bachelor of Computer Science @UCL

Los Angeles, California, United States

About

Experience

  • Machine Learning Engineer at Moloco
    Nov 2025 - Present · 9 mos

  • Machine Learning Engineer Intern at Moloco
    Jun 2025 - Sep 2025 · 4 mos

  • Research Assistant at UCL
    May 2023 - Jun 2024 · 1 yr 2 mos

    Collaborated with clinicians from UCL Hospital to design and deploy machine learning models, including RNN and XGBoost, in Python to predict patients’ biological age using both time-series and static clinical data. Integrated the models into a production-like clinical decision support system, which achieved an average user rating of 4.8/5.0 from over 500 clinicians. • Developed data-processing and storage pipelines in Python using PostgreSQL and TimescaleDB, handling terabyte-scale time-series data collected from over 200 patients’ wearable monitoring devices. Optimized distributed data processing to improve training throughput and reduce hardware costs by £13,000. • Implemented a Kafka and Spark Streaming system to support hourly online and offline training of a Generative Adversarial Network model. Achieved an accuracy of 95%, replacing traditional manual data completion methods and reducing the website data iteration time to minutes.

  • Course Representitive at UCL
    Oct 2021 - Sep 2022 · 1 yr

    I work as a communication channel between staff in our department and students.

  • Research Assistant Intern at UCL Hawkes Institute
    Jun 2022 - Aug 2022 · 3 mos

    Developed a Windows desktop application using C++ and Python that quantitatively evaluates the effect of the defogging model on downstream segmentation tasks. It has been widely used in the model development tasks of 15 research groups, improving the iteration efficiency of the models by 50% through the automation of the model evaluation process. • Utilized the gRPC communication protocol to support efficient communication between the C++ client and the Python backend GPU server, resulting in a 10% improvement in the software’s response speed.