Selected Projects

Software, AI, data visualization, and systems projects from my portfolio.

What's This?

These projects show the range of systems I like building: full-stack applications, AI workflows, data dashboards, automated content platforms, and lower-level storage infrastructure. I update this page as projects evolve. If you have questions, email me at kevin.you@mail.utoronto.ca.


1-on-1 - Event Planner Application

  • Engineered a microservices backend with Node.js and Express.js, reducing average API response latency from 400ms to 300ms and supporting 200+ concurrent users during peak event planning.
  • Designed a scalable RESTful API with Django REST Framework, cutting critical data retrieval time from 300ms to 210ms for user event queries.
  • Implemented real-time chat with WebSocket and Socket.IO, achieving message delivery latency under 100ms and a 20% improvement over the first prototype.

HotSpot - Residential Application

  • Contributed to a group project with TD Management Data and Analytics Lab and received Best Visualization Award in UofT's DataFest.
  • Built an interactive full-stack dashboard for commercial lease transaction data, enabling users to explore 10,000+ historical data points with filters and live visual updates.
  • Developed reusable React visualization components that improved user engagement by 30%.

Younee AI - Personalized Outfit Recommendation Engine

  • Engineered a full-stack application that provided personalized outfit recommendations to 500+ users by analyzing style preferences, weather, and fashion trends.
  • Developed RESTful APIs with Spring Boot, reducing average API response latency to less than 200ms while managing 1,000+ user profiles and outfit records in MySQL.
  • Integrated GLM-4-Flash into the recommendation engine, improving suggestion relevance by 25% based on user feedback metrics.

TDS News AI Journalist - Automated Content Generation System

  • Engineered a scraping pipeline with BeautifulSoup to collect and structure news from 50+ sources, building a curated dataset of 1,000+ articles and press releases.
  • Developed an AI journalist agent using GPT-4 to automate article generation from collected data.
  • Designed 10+ scalable RESTful API endpoints with Django REST Framework for frontend integration and automated publishing workflows.
  • Containerized the application with Docker to keep development and deployment environments consistent across the team.

National Pollutant Release Inventory Visual Report

  • Developed a full-stack interactive web platform to host and communicate environmental data analytics.
  • Built and deployed an R Shiny facility map that lets users explore emissions for 100+ toxic substances across 200+ facilities and all Canadian provinces.
  • Transformed Canadian National Pollutant Release Inventory data into actionable visual insights for policymakers and the public.

MimicDB - High-Performance Key-Value Store

  • Architected a scalable LSM-tree-based key-value store, using an AVL tree for in-memory operations and B-tree indexes for on-disk SSTables to boost query throughput by 60% over binary search.
  • Designed a 10MB buffer pool with a custom hash map and integrated Bloom filters to reduce unnecessary disk I/O.
  • Built and benchmarked the system with Makefile, reaching 1,700 ops/sec for updates and 800 ops/sec for searches through optimized compaction policies.

SDG Data Visualization Dashboard

  • Developed a Streamlit application to visualize UN SDG indicators with four analytical tabs: Map, Trend, Comparison, and Data.
  • Won Best Visualization Award in UTSC's SDG Challenge Competition.
  • Used Plotly to build animated choropleth maps and interactive charts, handling datasets with 50,000+ records without performance degradation.
  • Implemented a Pandas data layer for ETL, aggregation, and filtering across goals, targets, indicators, and time periods.