TAI :
Teaching Assistant Intelligence
for UC Berkeley
#LLM #Education Tech #Web Interface
Role
UX Research | UX/UI Design | Brand Design
Team
1 Project Lead
1 PM
3 UX/UI Designers
12+ Software Engineers
Timeline
Fall 2024 ~ Spring 2025 (Part-Time 5 months)
TAI is aimed to empower students, TAs, and professors with a personalized LLM-based Teaching Assistant that simplifies access to knowledge, fosters deeper learning, and drives collaborative innovation in education. Funded by Qualcomm, we envision all institutes adopting TAI, starting with the Berkeley community to redefine the future of education through AI.
Process
Highlights
Impact
92%
User Satisfaction Rate
-68%
TA Appointments
20+
Usability Tests
Feature 1 : General Course Inquiries
Students can ask any general questions that are relevant to anything within the selected course boundaries
Most asked questions are suggested in the dropdown
TAI provides relevant information related to the query across multiple course materials (PDFs, Lecture Recordings etc)
Feature 2 : Recorded Lectures
Students can watch recorded lectures with transcripts generated by TAI
Students can ask questions on the lecture (within the selected course context) to TAI while watching the recording
Students can jump to key chapters of the lecture, auto-generated by TAI
Feature 3 : PDF Material Inquiries
Students can flip through PDF materials with bookmarks marking key topics generated by TAI
Students can ask questions on the PDF file (within the selected course context) to TAI
Feature 4 : Accessibility for Visually Impaired
Students can record their queries via voice
TAI replies by narrating the generated content
research
User Research
Understanding the struggles of TAs
TA roles for Berkeley's STEM courses are notorious for their workload. Professors receive multiple requests for additional TA recruitment to support the workload, and students express their struggles with waiting to gain feedback from TAs.
We conducted in-depth user interviews on 15+ TAs of STEM courses to figure out their perspective.
Through our research we learned :
- 85.7% of STEM TAs support large, high-enrollment undergraduate courses
- TAs commit at least 20+ hours per week with additional extended hours on office hours, lab prep, and grading
- Repetitive Tasks & Inquiries lead to inefficient support for large number of students
"We are always so overwhelmed..
We can't get to every student"
Understanding the struggles of Undergraduate Students
Undergraduate students also have their struggles with their STEM coursework.
We interviewed 20+ undergraduate students and surveyed 70+ students enrolled in STEM courses.
Through our research we learned :
- 18% of STEM undergraduates didn't manage to have office hours with TAs due to the long waitlist
- Cross-referencing course resources for comprehensive understanding is difficult
- External LLM services provide overwhelming irrelevant information related to coursework
"I can't get help from TAs and ChatGPT keeps overwhelming me with too much advanced concepts.."
objective
How might we unburden workload for TAs and aid students with their comprehensive study across course material mediums?
Design
Design System
Further updates in progress..
Copyright©2024. All right reserved Sun-Q Kim
