Case study · Master’s thesis

Multilingual AI assistant for inclusive university portals

Why translation alone did not help international students, and what did.

Role: Lead researcher (thesis)
Context: Drexel University · funded by an IDEA-DEC MicroGrant · advisor Prof. Troy Finamore
Methods: Surveys (n=34), semi-structured interviews (n=9), IT support log analysis, task-based usability testing (n=15)
Participants: International students, including Farsi-, Spanish- and Chinese-speaking users
Team: Two-person build team (UI designer, web developer) that I hired and directed
Output: Master’s thesis (DOI), INTED2025 and EDULEARN25 papers

Context

International students at Drexel struggled with DrexelOne, the university portal. They could not find what they needed, abandoned tasks, and blamed themselves, even though the problem was the system. I went in assuming that translating the interface would fix it. The research showed otherwise.

Early research highlighted language barriers, information overload and onboarding stress.

Approach

I combined surveys and IT support logs to see where problems were common, then used semi-structured interviews and think-aloud task testing to understand why. I watched where students paused, where they gave up, and where they said “I don’t understand this.”

Workflow mapping of a common portal task.

Deep navigation, unclear labels and scrolling caused confusion.

What the surveys showed

More than two-thirds of respondents (68%) found it hard to locate essential features, especially registration, financial aid and billing.
Half described course registration as confusing and inefficient.
47% said the portal lacked the language support they needed, naming Persian, Spanish, Korean, Hindi and Gujarati.
Students relied on machine translation for financial and academic terms, and did not trust it.

“I have to use Google Translate, but the translations are inaccurate.”

From my analysis of 34 survey responses and 9 interviews.

Finding

Students rejected “robotic” translation. Word-for-word translation made terms more confusing; they needed concepts explained in language that felt culturally right.

This also surfaced dialect support as a key limit for trust: a translation could be technically correct and still feel foreign.

Design response

Instead of translating everything, we designed D.A.V.E., an AI assistant that explains tasks in context, with tiered language support and adaptive navigation. We also flattened navigation and rewrote labels so the portal became clearer for everyone, not only international students. Testing showed improved task success and less stress for participants.

Early D.A.V.E. concepts focused on simplified language, personalization and guided help.

Prototype

Walkthrough of the thesis prototype and the research behind it.

What stayed with me

AI can help people feel less lost, but only if we first understand what is confusing them, in their own words and their own language. That question now shapes my PhD direction.