This abstract was part of the 2026 UCSF AI and Education Symposium
Faculty burnout is often driven by the repetitive task of creating new learning materials from scratch. Yet, departments sit on goldmines of existing content—years of recorded lectures and validated textbooks—that go unused because they are hard to access. What if an AI could watch your previous lectures, read your textbooks, and automatically draft a new, interactive slide session for next week's class?
In this 60-minute workshop, we will demonstrate the UCSF Knowledge Hub Pipeline, a system that does exactly that. Participants will learn how to transform their static files into an active "AI Teaching Assistant" that can synthesize new curriculum materials on demand.
Learning Objectives
- Digitize: Explain the process of converting video and PDF assets into machine-readable JSON formats using multimodal AI.
- Structure: Understand how to organize raw text into a hierarchical "Knowledge Graph" that respects medical taxonomy.
- Synthesize: Use a vector database to automatically generate quiz questions and slide decks from existing content.
- Verify: Learn to use "Source Linking" interfaces to check AI outputs against the original ground truth.
Active Engagement & Lesson Plan (60 Minutes)
00-20 min Part 1: The Input (Getting Data In)
We will break down the open-source pipeline available on GitHub. We will show how tools like OpenAI Whisper and Gemini 3 Pro "watch" video lectures to create data, and how that data is structured into a Knowledge Graph.
20-40 min Part 2: The Output (Generating Content)
This is the core "Teaching Assistant" demo. Using our Cytopathology Tool, we will select a raw transcript and use the AI to automatically draft a new slide deck.
40-50 min Part 3: Verification (Checking the Work)
We will demonstrate the General Pathology Prototype to show how we verify AI outputs. Participants will see how to "Click-to-Source" to ensure the generated content matches the original textbook or lecture.
50-60 min Q&A & Implementation: Discussion on how other departments can deploy this code.
Submitted by: Charlie Herndon, DO PGY-1 Department of Pathology and Laboratory Medicine