AI DRG Project
As part of my Directed Research Group (DRG) focused on prototyping with AI, I explored how to create equitable group formation processes by leveraging guideline-driven input and prompt engineering. This project highlights the potential of AI in addressing collaboration challenges and presents a scalable framework for forming balanced, diverse, and effective teams.
View the full prototype here.
Group work is a core component of HCDE (Human-Centered Design & Engineering) classes with heavy collaboration between product managers, designers, and researchers from diverse backgrounds. Currently, students either self-select their groups or instructors manually form teams based on student surveys. However, this often results in imbalanced teams, hindered collaboration or increased time demands for both students and instructors.
By conducting a survey and user interviews, I uncovered students' pain-points and needs in the current process of finding teammates through self-selection.
To tackle the main challenge of imbalanced teams and lengthy process of group formation, we decided to leverage AI to automate the process, delivering data-driven results that save time for both students and instructors.
Previous research showed that students heavily rely on digital collaboration tools for group work.
I explored three platforms most commonly used by HCDE students: Slack, Figma, and Canvas.
Slack would automate group chats for seamless collaboration but lacks design and project management tools. Figma could form teams based on intro cards with skills and goals, ensuring transparency but requiring extra setup. Canvas would leverage existing course data for easy group formation but offer fewer collaboration features.
I chose to explore Concept 1 (Slack Integration) to leverage the opportunity of providing a centralized communication hub. Slack also integrates with a variety of platforms for project management, design, and research, enhancing collaboration across workflows.
AI Prompt Engineering
Exploring prompts to guide design direction
I experimented with various unstructured data inputs in ChatGPT such as conversational text and LinkedIn content to test its ability to form meaningful clusters. By assigning different weights to parameters based on their importance, I prioritized qualities students found most important.
You can check out the final ChatGPT prompting process here.
AI Prompt Optimization
Designing a smarter team matching process with balanced data inputs
I chose resumes, multi-select checkboxes, and free-text input to prevent data overload and improve grouping accuracy. The balance of structured and unstructured data provides a more nuanced understanding of each individual’s skills and preferences while each method processes smaller and focused data sets.
Final Designs
01. A streamlined system in place for students and instructors
I designed two key entry points for group formation that cater to students' needs and preferences while aligning with the class syllabus requirements set by the instructor. This ensures that groups adhere to project guidelines.
02. Custom team matching through distinct profiles
Students complete a profile setup form, allowing AI to extract key information and keywords to build individual profiles. These profiles are then used to cluster students into groups based on shared goals, preferences, and complementary differences.
03. Team formation rationale for trust and transparency
Students receive AI-powered team recommendations and suggested project topic ideas from a Slackbot, along with a clear rationale explaining how each team was formed based on factors like skills, availability, and collaboration styles.
Learnings & Future Opportunities
By using prompt engineering with ChatGPT to understand the backend processes of AI, I was able to guide my design decisions to achieve more accurate results. While extensive collaboration with software engineers would still be needed to implement these designs in real life, this experience taught me how AI can effectively navigate and influence the design process.
Moreover, I worked with Slack’s existing design systems to ensure consistency, creating new components for features like profile and project forms. I developed these components to align with Slack’s visual identity and enhance the user experience, while supporting the addition of new features.
To further scale the group formation feature, I would introduce a parameter-adjusting slider system to dynamically fine-tune weighting, making team matches more personalized to each class and student’s needs. Additionally, I would explore edge cases such as handling conflicts when users are unhappy with their matches, to add flexibility and enable smarter adjustments. These improvements would make the team formation process not just more efficient, but also a better experience for everyone involved.