Episode Details

• Date: 2026-02-16
• Duration: ~7 minutes
• Speakers: Leo Zhadanovsky (Host, Enterprise Technologist for Education at AWS), Gayathri Rengarajan (Guest, Associate Director of Data Science at PowerSchool)

Episode Summary

In this special re:Invent episode, Gayathri from PowerSchool shares the journey of building Power Buddy, an AI-powered learning assistant that evolved from a hackathon project to a full suite of products serving over 60 million K-12 students. The conversation focuses on implementing robust content filtering using fine-tuned models on AWS SageMaker, addressing student safety concerns while reducing false positives, and the future of agentic AI in education technology.

Key Discussion Points

• Evolution of Power Buddy from hackathon project to production suite
• Implementation of strict content filtering for student safety in AI interactions
• Challenge of balancing safety with reducing false positive notifications for administrators
• Fine-tuning smaller models (Llama 8B) for domain-specific content filtering

Featured Technologies

• Llama 8 billion parameter model
• AWS SageMaker AI

Key Takeaways

• Fine-tuning smaller models for specific educational use cases reduced false positive rates to less than 3%
• Technology stack and models should remain flexible while keeping security, reliability, and scalability constant
• Close collaboration with cloud providers accelerates AI implementation and optimization
• The future of EdTech AI is moving toward agentic systems with improved observability and evaluation capabilities

Resources Mentioned

• AWS SageMaker AI
• Llama model family
• AWS Agent Core
• AWS Bedrock

Tags

#EdTech #K12Education #AWS #AI #MachineLearning #StudentSafety #ContentFiltering #PowerSchool #SageMaker #AgenticAI #reInvent