An informal venue for sharing ideas about harnessing AI — and addressing the issues it raises — in teaching, research, and administration within the Jacobs School.
If you’d like to present in the future, please fill out this form.
If you have any questions, please contact me at sjswanson@ucsd.edu.
Tech support provided by Mauricio Villasenor.
In keeping with the purpose of the meeting, this webpage was generated mostly with AI. Please excuse any errors.
Writing PrairieLearn Questions and Computer-Based Testing Practices
Dr. Zilles demonstrated PrairieLearn, a platform that supports parameterized question generators, automatic grading, and a browser-based exam environment built on an existing library of question elements. He covered mixed auto/manual grading workflows, an emerging AI-assisted grading feature, and a “no code” interface for authoring questions, and described how his group has used these tools to shift homework and exams from paper to fully computer-based formats.
Google Teachable Machine and Quick, Draw! in MAE 2
Dr. Sadeghizadeh introduced AI literacy to MAE 2 freshmen through hands-on activities with Quick, Draw! and Teachable Machine, sparking genuine reflection on AI’s limitations and the irreplaceable role of human judgment in engineering.
Identifying students-in-need with ChatGPT
Dr. Van Den Einde uses LLMs to analyze weekly student reflections and assignment data in a 350-student freshman engineering course, automatically drafting personalized instructor emails to identify struggling students early and make them feel seen — with a live controlled study underway this quarter.
Grading when students use GenAI for everything
Dr. Politz describes a grading system built around in-person “skill demonstrations” — proctored, step-by-step practical exams with coarse pass/fail rubrics and multiple retries — designed to give meaningful assessment of what students can actually do independently, in an era where collaboration, helpful TAs, and LLMs all confound take-home work.
Human-centered teaching
Dr. Soosai Raj redesigned a 450-student intro programming course around handwritten, process-oriented homework with manual grading and small student study groups — deliberately putting humans back in the loop at every stage to counter the hollowing-out effect of AI tools on authentic learning and student community.
AI Oral Assessment System
Dr. Wang built a custom AI-powered oral assessment tool that conducts structured voice interviews with students using an instructor-defined question bank and rubric — designed to restore authentic knowledge verification in an era where written reflections and paper summaries are trivially AI-generated.
A Flexible PyTest-to-PrairieLearn Code Auto-Grading Translation Layer
Dr. Moshiri presented a PyTest-to-PrairieLearn translation layer that makes it easy to author and deploy code auto-graders on PrairieLearn without rewriting grading logic from scratch. The layer consists of a Python parser that converts PyTest output into PrairieLearn’s required format and a companion shell script, demonstrated across three courses with custom Docker containers. Once graders are written, they enable fully automated, proctored, randomized coding exams at the Triton Testing Center, shifting course staff effort from grading toward improving grader quality and student engagement.
Using AI in a Research Group
Dr. Yu described a RAG-based Slack chatbot her lab built to reduce the repetitive informational burden of running a large research group of roughly 32 students. The bot indexes the lab’s GitHub wiki — containing onboarding documents, policy guidelines, server usage instructions, and other group knowledge — converts content into vector embeddings, and answers student questions directly in Slack around the clock. It can also query live GPU and TPU cluster status, helping students optimize experiment scheduling near deadlines with essentially zero ongoing maintenance.
PrePostClass: An In-Class Formative Assessment Platform
Dr. Khojah demonstrated PrePostClass, a web app for in-class formative assessment that administers short pre- and post-lecture quizzes to measure learning gains within a single class session. The tool uses LLMs to generate questions from uploaded lecture materials with a non-binary grading scheme that distinguishes guessing from genuine understanding, and integrates directly with Canvas for grade passback. Response data exports by Bloom’s taxonomy level, and participation scores correlated with final exam performance better than homework grades.
Design Lab Faculty Intelligence
Dr. Weibel presented three interconnected tools built to automatically track and surface the activities of the Design Lab’s 50+ faculty across 20 departments without requiring faculty to self-report. A monthly pipeline scrapes and verifies faculty profiles from public sources, feeds a highlights digest for the lab newsletter, and powers a Synergies Finder that matches incoming opportunities — RFPs, funding calls, delegation visits — against the faculty database to rank the most relevant faculty. He demonstrated the system against a Singapore University delegation visit and an NSF grant opportunity.
Automating Tedious Web Tasks
Prof. Swanson described using Claude Code to build locally-running Chrome extensions that scrape data from university internal websites without sending screenshots or page content to the cloud. He demonstrated a tool for UCSD’s SETS course evaluation system that reverse-engineers the site’s internal API calls and provides a simple interface for querying evaluations by department, course, or faculty name and downloading results as a spreadsheet. The approach generalizes quickly to other internal university sites, though it requires some care to avoid inadvertently overloading backend APIs.