Tech Stack
Description
Drawing from my experience teaching English in Vietnamese kindergartens, I wanted to build a tool that helps ESL teachers generate age-appropriate lesson materials without spending hours searching for resources. Teachers working with young learners (ages 0-6) need songs, vocabulary, games, and activities tailored to specific age groups and proficiency levels - and everything needs to reinforce the same vocabulary for effective learning.
I'm building a Next.js application where teachers input age range, English level, and lesson topic to receive a complete set of teaching resources. The system uses the Gemini API with custom prompts informed by my teaching experience to generate songs with embedded YouTube videos from trusted ESL channels, age-appropriate vocabulary lists, interactive games with variations and required materials, and speaking prompts with expected student responses. The prompt architecture adjusts resource complexity and quantity based on age and proficiency - for example, advanced 5-6 year-olds receive 7 vocabulary words versus 5 for beginners.
The application currently generates and displays complete lesson plans with working YouTube embeds. This is a personal project in active development as I explore LLM integration patterns and prompt engineering for educational tools.
Key Features
- Developed modular prompt system with base role definition and age-specific templates (0-3 years, 4-5 years, 5-6 years) based on early childhood education principles
- Built dynamic resource allocation that adjusts quantities and complexity based on age and proficiency level intersection
- Integrated Gemini API with structured JSON response parsing for consistent data handling
- Implemented YouTube Data API integration that searches for generated song titles and embeds videos directly in lesson plans
- Designed prompts incorporating developmental appropriateness, vocabulary reinforcement across sections, and scaffolding from sensory activities to verbal production
- Created authentication system with planned expansion for saved lesson libraries and custom resource uploads
Current Focus
- Performance optimization: implementing streaming responses and caching to reduce generation time
- Enhanced AI architecture: refactoring to use function calling and tool use for more efficient responses
- User features: Expanding authentication to support saved lesson libraries, custom resource uploads, and a dynamic lesson plan builder where teachers can mix AI-generated and manual content
- Community features: Plan sharing and rating system for teacher collaboration
- Additional resources: AI-generated flashcards with export options for printing or classroom display
- Codebase migration: Converting from JavaScript to TypeScript for better type safety
- Testing implementation: Adding Jest unit tests and Cypress end-to-end tests