How might university students use an AI-powered platform to find compatible study groups so that they can reduce academic stress and improve their overall learning experience?
Why finding the right study group is harder than it should be, and who pays the price.
As students, study groups are a huge part of the college experience, whether inside the iSchool or not, especially for hard classes and collaborative activities. The main issue that remains is: how do you find a group that works for you? Classes, clubs, where you live, your personality, your preferred learning type (visual, auditory, or kinaesthetic); so many factors come into play when trying to optimize your experience.
It becomes really hard to do so, and the consequences are real: students end up with worse grades, higher stress, and that familiar feeling after finals of "I wish I had found a better group for X class." Beyond individual grades, poor study group experiences chip away at students' sense of belonging and confidence on campus, one of the biggest drivers of students dropping out or feeling disconnected from their university.
This issue affects every student on campus, but especially first-year students, transfer students, and anyone jumping into a large, high-stakes course without an existing social network to lean on.
Our solution aims to leverage AI-based algorithms to help optimize forming study groups using details such as time schedules, clubs, personalities, preferred learning type, and more, pairing students into class-wise study groups to optimize the learning experience that university has to offer. We want to make the process of forming study groups more accessible for the entire UW student community, and possibly other schools as well, with a strong focus on accessibility so the platform serves students who are most often left out.
Tools like GroupMe, When2Meet, and Discord let students communicate but don't intelligently match them before the conversation starts, leaving the most isolated students completely underserved.
Students face academic pressure at university and struggle to find study groups, with the current process being largely random and inaccessible.
First-years and commuter students without existing social networks are left studying alone or stuck in mismatched groups that don't fit their schedule, learning style, or personality.
There is no centralized way to match students based on what actually matters: class enrollment, availability, learning type, and personality, so students resort to luck or word of mouth.
Tools like GroupMe, When2Meet, and Discord let students communicate but don't intelligently match them before the conversation starts.
An AI-powered platform that takes in class enrollment, availability, learning type, personality, and club involvement to intelligently pair university students into compatible study groups.
Our problem is an adaptive challenge. While technical challenges may arise when building an AI matching platform, the harder problem exists within the behavioral and social realm. Students must be willing to share personal details like personality and learning style, trust an algorithm to find them a group, and then actually show up and engage with strangers. No amount of good engineering solves someone who is too anxious to join, too skeptical to fill out a profile, or too set in their habits to try a new way of finding a study group. Introverts remain a hard target audience. The solution requires changing how students think about and approach group formation entirely. This is a social and behavioral shift, not just a technical one.
We employed a concurrent mixed methods design: a quantitative survey (10 respondents from a fraternity GroupMe) and qualitative semi-structured interviews (3 freshman students). Participants varied in class standing, 2 Seniors, 4 Juniors, 2 Sophomores, 2 Freshmen.
Students self-sort into subject channels. Peer learning thrives on platforms students already trust, but group quality and compatibility are inconsistent and unmoderated.
Pairs students by course, schedule, and learning preferences via swipe UI. Lowers browsing friction but doesn't address the behavioral gap of actually showing up.
Bar exam prep for law students. Highly niche but demonstrates deeply tailored matching logic for one learning context.
Scholarly research reinforced that study groups develop critical thinking and listening skills difficult to achieve alone (Linn et al., 2013), while post-COVID social anxiety makes proactive outreach harder for many students (Khuda et al., 2024). A systematic review of 27 studies found AI recommender algorithms improve collaborative learning outcomes through personalized group formation (Kovari, 2025), and working implementations like BuddyIN demonstrate strong matching performance using course enrollment and personality traits (Ko, 2025). Revelo Sánchez et al. (2021) showed Big Five-based genetic algorithm grouping significantly outperformed student self-selection on post-test grades, directly supporting our personality-first approach.
Key decisions grounded in research, values, and the trade-offs we accepted.
We chose an AI-powered web platform because our research showed students do not lack communication tools; they lack intelligent, trust-building matching before the social risk of reaching out. StudyHub combines course-aware matching, personality data, transparent reasoning, and structured first sessions in one place, rather than asking anxious students to self-organize on Discord or swipe through profiles without context.
Unlike Discord, StudyHub moderates compatibility rather than leaving students to self-sort in chaotic channels. Unlike Sidekick, it addresses the behavioral gap of showing up, not just browsing matches. Unlike Studicata, it targets the general undergraduate population at UW while still allowing course-specific depth. Compared to word-of-mouth group formation, it proactively reaches students without existing networks.
Interviews and surveys directly shaped our feature priorities: profile preview before committing, reputation-driven trust signals, plain-language match explanations, and introvert-first onboarding. Usability testing further revealed that personality test discoverability and schedule setup are make-or-break moments, driving our redesign priorities.
Students consistently rated personality alongside or above academic factors. Bad group experiences were described as personality mismatches, not just academic ones.
Because: Research showed personality fit is as important as academic alignment.
Trade-off: More profile data raises surveillance concerns; we must be transparent about how matches are made.
Students need to feel safe before committing to a stranger. Seeing profiles before joining was selected by 6/9 survey respondents.
Because: Familiarity reduces the risk of a bad match; we surface enough info to build trust without overwhelming.
Trade-off: More visibility vs. privacy; we balance transparency with data protection (value-sensitive design).
Students default to people they already know because it feels safe. An AI matcher asks students to trust a system over their own instincts.
Because: Literature (García-Vélez et al., 2021; Revelo Sánchez et al., 2021) shows data-driven matching outperforms self-organized groups.
Trade-off: Too much user control defeats the purpose; too little feels like lost autonomy. Designing the handoff is critical.
The core challenge is not technical but behavioral. Algorithmic matching only creates value once students feel safe enough to share personal details and act on a recommendation.
Because: Affinity mapping surfaced that introverts and anxious students need alternative entry points that don't require initiating contact.
Trade-off: Optimizing for anxious users may feel slower for extroverts who prefer spontaneous group formation.
Our affinity mapping session organized interview and survey data into user needs, recurring problems, and design opportunities. This synthesis served a similar role to an empathy map by centering student emotions around safety, trust, and social risk.
Because the platform serves the UW community, we prioritized data protection while ensuring a smooth experience. Personality is the primary driver in group formation, prioritizing human compatibility and wellbeing over pure academic efficiency. A tension exists between user autonomy and algorithmic optimization: students default to people they already know because it feels safe. Designing the handoff without making users feel they've lost control is a challenge we continuously account for.
Core design decisions (personality-based matching, profile visibility before committing, and reputation-driven trust signals) were surfaced by students themselves and validated across both interview and survey methods. Our participants acted as informants rather than co-designers, and our sample was relatively homogeneous (largely freshman males), which limits how fully our design represents the diversity of the UW student population.
StudyHub addresses the accessibility gap in collaborative learning through trust-building onboarding, personality-aware matching, and tools for group follow-through.
Think-aloud usability testing with Wizard of Oz matching, two rounds of prototype testing, and Nielsen heuristic evaluation.
What we learned throughout the design process and how it changed our approach.
Throughout this project, our team learned a great deal about the design process, especially since many of us had limited experience with Figma, design prototyping, and user testing at the start. Despite this, we were able to develop a functional and thoughtful design by continuously incorporating peer feedback and applying what we learned throughout the project. Feedback from classmates and users helped us identify areas for improvement, such as making buttons more visible, simplifying navigation, making pages less cluttered, and creating a more intuitive user flow.
One of the biggest challenges our team faced was coordinating our work schedules. Since many team members had other commitments, it was sometimes difficult to find time to meet, communicate effectively, and delegate tasks. Another challenge was learning the technical aspects of Figma while simultaneously designing the prototype. As we progressed, we became more comfortable with the platform and gained a better understanding of how to create interactive and user-friendly designs.
The design and testing process taught us valuable lessons about user-centered design. We learned how to conduct more effective user testing sessions by asking clearer and more targeted questions that generated useful feedback. We also gained practical experience in translating user needs into design decisions, for example, shifting toward personality-first matching and making the personality test entry point more prominent after usability testing revealed it as a critical failure point.
If we had more time, we would expand the platform by implementing additional features that were part of our original vision but could not be fully explored during this project. For example, we would like to add collaborative tools such as a shared whiteboard space and group streaks to encourage engagement and teamwork. We would also run co-design workshops with a broader range of students beyond our relatively homogeneous research sample.
Looking ahead, our design would need to continue evolving to meet the changing needs of users. As the platform grows, we could expand its scope to support additional communities, such as Registered Student Organizations (RSOs), research groups, and other collaborative academic communities. By continuing to gather feedback and adapt to user needs, StudyHub could develop into a more comprehensive and versatile tool that supports a wider range of users and activities.
APA 7th edition