Phone: 878-218-8239
Email: changyiyang2023@gmail.com
GitHub: github.com/ChangyiYang
Education: Carnegie Mellon University
Carnegie Mellon University
Information Networking Institute
Aug. 2024 - Present
University of California, Berkeley
Exchange Student in Computer Science
Sep. 2022 - May. 2023
GPA: 3.96/4.00
Chinese University of Hong Kong, Shenzhen
Bachelor of Science in Computer Science
Sep. 2020 - June 2024
GPA: 3.92/4.00
Long-Short Term Memory (LSTM) for Pebble Bed Reactor
UC Berkeley, California
Jan. 2023 - Jan. 2024
- Applied Data Science and Machine Learning techniques to the nuclear reactor field, providing ML insights for physicists.
- Used LSTM to predict the reactor core state, helping the operator better control the reactor and avoid accidental mistakes.
- Designed an effective dimensionality reduction method that had physical meaning, resulting in 1000 times size reduction.
- Used Variational Autoencoder (VAE) to do data duplication, solving the insufficient data issue. Expanded the data by fivefold.
Sony
Shenzhen, China
Backend Engineer Intern
Jan. 2024 - May. 2024
- Integrated data from IMU sensors and MediaPipe's posture recognition model using Kalman filtering to enhance motion tracking accuracy, particularly for complex hand and torso movements during slow motion.
- Balanced speed and accuracy by selecting appropriate model and image quality based on the online and offline scenarios.
- Built a multi-thread video frame capture system using OpenCV to achieve 60FPS 1080p real-time frame capture.
- Addressed IMU drift by using MediaPipe's posture estimates to reset the torso's angle when the subject faced forward.
ByteDance (Tiktok’s parent company)
Beijing, China
Backend Engineer for AI Lab Smart Audio Team
May. 2023 - Aug. 2023
- Optimized audio data handling (50TB/day) for Douyin by reducing redundant data storage using Magnus, an Iceberg Datalake framework’s git-like branch feature, enabling label isolation across teams and greatly reducing storage cost.
- Switched data storage format to Parquet, a columnar storage format to support data flow into Magnus datalake.
- Designed a partitioning strategy to bucket data by time, reducing query costs for view tables for different teams.
- Migrated the speech processing workflow to a new platform, completing key worker registration, XML configuration, node testing, and workflow optimization. Modified the worker to suit user-defined parameters and DAG workflow scheduling.
FortuneDraw Points Lottery System
SpringBoot, MyBatis, MySQL, Redis, SpringCloud
- Built the lottery system using Domain-Driven Design with multiple design patterns, ensuring maintainability and scalability.
- Performed performance testing using JMeter to identify and optimize high-latency APIs, improving throughput by 50%.
- Designed the lottery standard with Template Method pattern, applying the Chain of Responsibility to execute the draw.
- Implemented a dynamic decision tree with Composite pattern for strategies after draw, supporting configuration via database.
- Optimized inventory control for high-traffic using Redis with async queues and scheduled updates to reduce database load.
Dynamic Thread Pool
SpringBoot, Nacos, Prometheus, Grafana
- Integrated Nacos for centralized configuration, enabling unified management and dynamic updates of thread pool parameters.
- Created a custom Actuator Endpoint class to manually expose thread pool metrics, integrated with HertzBeat for monitoring.
- Configured Prometheus alert rules and Grafana dashboards to enable real-time notifications for thread pool anomalies.
- Leveraged the SPI mechanism to allow users to customize and extend thread pool rejection policies.
DishNow Food Delivery Platform
SpringBoot, Spring Security, MySQL, Redis, MyBatis, Nginx
- Implemented authentication using Spring Security and JWT, with an RBAC model for permission management.
- Used AOP to automatically populate common fields (e.g., creation time, creator), reducing code and coupling.
- Used Bloom filters to prevent cache penetration and SpringTask for timed cache pre-warming to avoid cache avalanche.
Microsoft Fabric and AI Learning Hackathon
- Designed and implemented an automated system to collect and process images based on user-defined parameters, ensuring consistency in format and dimensions through automated format conversion, super-resolution, and cropping on Azure.
- Integrated data pipelines and storage triggers to automate the entire image life cycle, from collection to packaging.
- Applied advanced ML models to enhance image resolution and perform precise cropping, improving image quality.
- Collaborated with two MLEs on model selection, deployment, and testing, seamlessly integrating models into the data pipeline.
- Programming Languages: Proficient in Java, Python. Intermediate in Javascript/TypeScript, C++. Basic knowledge of Go.
- Frameworks: Spring Boot, Spring Cloud, MyBatis, Hibernate, React, RESTful API Design.
- Databases: MySQL, Redis, PostgreSQL, MongoDB, SQL (Proficient).
- Cloud & DevOps Tools: Docker, Kubernetes (K8s), AWS (EC2, S3, RDS), Nginx, Jenkins, Git, Maven.
- Microservices & Distributed Systems: Spring Cloud, Kafka, RabbitMQ, Nacos.
- Testing & Monitoring: JUnit, Mockito, Postman, Prometheus, Grafana.