David YangFullstack product engineer

Platform that corroborates business registration data with health inspections and permits to identify new restaurant openings before they open. Developed with Next.js frontend hosted on Verceland FastAPI backend deployed on a DigitalOcean VPS.

Personalized global trip itinerary generator powered by the Claude 3.5 Haiku model from Anthropic. Interactive chat interface allows users to learn about their destination and view notable activities, flights, restaurants, and weather. Implemented an MCP server to enable the LLM agent to call tools when needed.

Hybrid image captioning model combining a CNN image encoder (ResNet-50) with a Transformer text decoder to generate natural language descriptions of images. Trained on the MSFT COCO dataset, on a cloud Nvidia H100 GPU from RunPod. Model hosted on Hugging Face.

Tool that allows students to upload course materials and notes to get personalized study materials including spaced repetition flashcards and practice quizzes. Developed with Next.js frontend and FastAPI backend deployed on a DigitalOcean VPS.

Reinforcement Learning agent trained on over 20 years of historical market data for ETFs: SPY, QQQ, VOO, and Tech Stocks: AAPL, GOOG, MSFT. Explored the application of various machine learning models including classification and regression before trying model free reinforcement learning with Q-learning and Deep Q-learning. Defined a reward function that optimized for a balance between portfolio profit and risk management.

Swift-based iOS app that leverages the GPT-3.5-turbo model from OpenAI to generate personalized road trip itineraries. Features Firebase auth and user data persistence.