Edison 3.0 is the newest iteration of an exciting educational AI system. Edison 3.0 is a multimodal Retrieval-Augmented Generation (RAG) system that handles a wide range of student questions, from complex mathematical concepts and programming questions to course-specific assignments and logistical inquiries. Unlike traditional educational chatbots, Edison combines advanced functional calling and hierarchical retrieval mechanisms to generate pedagogically aligned responses with TAs in the loop.

At its core, Edison features a modular architecture that processes both text and visual content, enabling students to submit questions containing equations, code screenshots, and diagrams. The system’s retrieval module simultaneously searches across course materials and historical Q&A databases to provide comprehensive context for response generation.

Edison has been successfully deployed in UC Berkeley’s Data 100 course, serving over 1,200 students across multiple semesters. This large-scale deployment provides a real-world testing ground where Edison handles everything from clarifying statistical concepts to debugging Pandas code and explaining machine learning algorithms. The system seamlessly integrates with the course’s EdSTEM discussion forum, allowing students to opt in to AI assistance while maintaining full TA oversight and quality control.

Our extensive evaluation demonstrates Edison’s remarkable effectiveness: 84% of responses require minimal or no instructor editing. In this demo, we’ll showcase Edison’s multimodal capabilities across mathematics, computer science, and data science domains, demonstrating how its modular architecture enables rapid deployment and customization for different educational contexts while maintaining instructional effectiveness in answering student questions.