Projects

Twitch+: Personalized Recommendation Engine

• Technologies: Spring Boot, React, MySQL, AWS RDS, OpenFeign
• Description: Full-stack application for personalized Twitch resource recommendations
• Key Achievements:
  • Enhanced user engagement by 25% through personalized content recommendations
  • Implemented RESTful APIs to fetch real-time Twitch resources
  • Hosted MySQL database on AWS RDS with efficient CRUD operations
  • Developed content-based recommendation algorithm
  • Containerized and deployed to AWS App Runner for scalable service

Email Helper: AI-Powered Email Generation

• Technologies: Python Flask, OpenAI GPT-3.5, Langchain, AWS DynamoDB, AWS Lambda
• Description: Conversational web app for generating email scripts using AI
• Key Achievements:
  • Integrated OpenAI's GPT-3.5 Turbo and Langchain for sophisticated AI-driven communication
  • Hosted NoSQL DynamoDB for email history management
  • Performed ETL tasks on Databricks
  • Containerized application on AWS ECR and deployed to AWS App Runner
  • Automated testing, containerization, and deployment using GitHub CI/CD

Plant Forager Mobile App

• Technologies: React Native, Expo, Node.js, Express.js
• Description: Cross-platform mobile application for plant discovery and personal plant repository
• Key Features:
  • Designed nature-themed UI using Figma
  • Implemented RESTful APIs for user profiles and plant collections
  • Utilized native mobile functionalities:
    • Location services for plant locating
    • Camera for plant profiling
    • Calendar and push notifications for plant ripening alerts

Incentivized Truthful Communication for Federated Bandits

• Publication: Co-author, ICLR 2024 Journal
• Research Focus: Federated Learning and Incentive Mechanisms
• Key Contributions:
  • Developed an innovative incentive-compatible communication protocol for federated bandit learning
  • Theoretically proved the model's sub-linear regret and communication cost
  • Designed and conducted comprehensive numerical experiments in multi-client federated bandit settings
  • Challenged existing assumptions about client truthfulness in distributed learning systems

Graduate Program Recommendation System

• Technologies: TensorFlow, Scikit-learn, Word2Vec
• Project Scope: Machine Learning， Recommendation System
• Technical Achievements:
  • Engineered a sophisticated recommendation system using application attribute analysis
  • Preprocessed and cleaned a large dataset of 80,000 subject records using advanced data pipelines
  • Implemented advanced feature representation techniques:
    • Applied Word2Vec for converting categorical variables into dense vector representations
    • Significantly enhanced model's information extraction capabilities
  • Conducted comprehensive algorithm evaluation:
    • Compared multiple machine learning approaches including K-Nearest Neighbors and ResNet
    • Achieved 73% accuracy through meticulous hyperparameter tuning
    • Demonstrated expertise in model selection and optimization