• Design and build evaluation systems that assess educational AI quality across thousands of conversations - from learning outcomes to bias detection to curriculum alignment • Build ML systems that enable self-improving app creation - learning from high-quality apps on the platform to automatically scaffold new applications for educators • Design and prototype downloadable, on-device AI models that work without internet connectivity - critical for privacy and global accessibility • Develop systems that enable dynamic, fluid interfaces adapting to learning moments - transitioning seamlessly from chat to writing editor to interactive physics simulation as needed • Build content moderation and safety systems designed specifically for educational discourse • Implement agentic AI systems that enable educators to create goal-directed applications (e.g., "help students through this project over 2 weeks") • Build sophisticated RAG systems that integrate diverse educational content with semantic search and knowledge graphs