Instructional AI

UC San Diego's Instructional AI pilot helps participating instructors offer a course-grounded AI Tutor through TritonGPT. Instructors choose the learning purpose, approved source material, and expectations for student use.

The pilot uses retrieval-augmented generation to bring instructor-selected course material into a response at request time. Because generative AI can still be incomplete or wrong, students should verify important answers against course sources and instructor guidance.

Model access follows the pilot's approved UC-hosted or enterprise cloud route. The appropriate route and controls depend on the course, source material, and intended use.

Results reflect the 2026 course-pilot survey. The response count is not available, so these percentages should be read as bounded pilot feedback rather than a campus-wide result.

How It Works for Students

Each participating course can receive a student-facing AI Tutor configured for that course and made available through the approved course access path.

The AI Tutor is designed to help students engage more deeply with course materials:

• Acts as a Socratic tutor: Guides students toward answers and asks them to explain their reasoning.
• Grounded in course content: Retrieves from instructor-selected material while generating a response; students should still check the cited source.
• Configurable pedagogical style: Instructors choose between Socratic (guides toward answers) or directive (provides answers) modes.
• Self-paced support: When the pilot service is available, students can explore topics at their own pace.

Instructor Setup & Control

Instructors are the gatekeepers of their course AI Tutor. You control what the tutor knows and how it behaves:

• Select course materials: Choose the approved documents and websites the tutor may retrieve from. Availability and refresh timing depend on the configured source.
• Set the pedagogical style: Choose Socratic mode to guide students toward answers, or directive mode to provide direct explanations.
• Multiple tutors per course: Create separate tutors for different units, lab sections, or study purposes.
• Evaluate the pilot: Work with the program team to review appropriate aggregate evidence and student feedback.
• No student-inserted content: Students cannot add their own materials to the tutor's knowledge base — only instructor-approved content is used.

Getting Started

Participation begins with a course-level conversation about purpose, source material, student expectations, access, and evaluation.

1. Request participation: Email AI-Pilots@ucsd.edu to join the program. Our team will help you get set up.
2. Review the course setup: Identify the approved source material and confirm the access path and refresh behavior with the program team.
3. Configure your tutor: Select which documents the tutor can reference and choose your preferred pedagogical style (Socratic or directive).
4. Communicate expectations: Tell students where to access the tutor, what it is for, how to verify answers, and what course policies apply.

Models and source material

Queries use UC San Diego's approved model gateway. Source types and model selection are configured for the instructional use case and may change as the pilot is evaluated.

Support & Logistics

• Collaboration: TAs and co-instructors can be added to help manage the course AI Tutor.
• Canvas Integration: The AI Tutor appears in the Canvas course navigation for enrolled students.
• Evaluation: The program team and instructor agree on appropriate feedback and quality measures for the course.
• Support: Available support and response expectations are confirmed during onboarding.
• Onboarding: New instructors receive guided setup and documentation before the term begins.
• Data handling: Approved model routes and controls are selected for the course. Instructors should use only the source material approved for the pilot.

Join the Program

We are recruiting instructors for Summer Session 2 and Fall 2026. If you are interested in participating or want to learn more, email us at AI-Pilots@ucsd.edu to schedule an overview.