EllisShang

Case Study

AI-Powered Vocabulary Learning Education Website

AI Engineer, Full-stack Engineer · Spokabulary · Jul 2024 - Dec 2024

View live project

Overview

Built an AI-powered web platform using Next.js, Unity, and PyTorch that lets students create and play word games from custom lists, with multi-agent 3D characters and NLP-driven chatbots.

Key Technologies

PythonPyTorchNext.jsTypeScriptUnityWebGLReactOpenAI APIFigmaNode.js

Story & Process

Overview

Spokabulary is a web-based tool that lets you create and play word games from your own vocabulary lists. You can choose from different game modes, such as Speak and Spell Challenge or Out of Order. Each game can be customized with AI to create tailored word lists that help students memorize vocabulary more effectively.

In this role, I worked as an AI engineer and full-stack developer, combining deep learning, multi-agent 3D characters, and NLP chatbots with a modern web experience.

Responsibilities

  • Co-designed the overall web and AI architecture in collaboration with the product manager.
  • Designed the web UI/UX in Figma to create an experience tailored to elementary and middle school students in North America.
  • Developed both the frontend and backend using Next.js, and optimized Unity-based games with WebGL to improve page load performance.
  • Tuned multi-agent prompts and LLM chatbot behavior to adapt to different vocabulary levels and learning needs.

Impact

  • Deployed an AI-powered gamified learning platform used by students to practice vocabulary with interactive 3D characters and conversational agents.
  • From Oct 2024 to Jan 2025, the platform reached around 3,000 monthly active users.