CLIENT WORK

A next-generation AI college counselor


Brief
I was brought in to design a demo for an AI college counselor. Within four weeks, we’d shipped a minimum viable product, and used it to successfully raise seed capital for further development.

Team
I was the solo designer. Cross-functional team of PMs, AI engineers, full-stack devs, investors, and leadership

Time
4 weeks, 2023



The problem space: college counseling is critical for the journey to higher education, but it’s inaccessible and expensive. My stakeholders are betting that AI is critical to address this gap.




App features

What’s it like to use a chatbot that isn’t just chatGPT?



Getting started with chatting with AI and transcripts

Students are encouraged by AI-generated suggestions to start the conversation. The chatbot will try to know more about the student.





Making a personalized application plan

A progress bar shows current and future topics and allows users to define their plan. 
It’s also a user-centric method of steering an LLM to stick to a consistent user journey over a long period of time.






When talking about your application, the agent cites relevant information.

In a conversation, agents can tag your information with dynamic UI that links to your profile.
It makes responses from AI more than just a wall of text.
   


You can view and edit this information, making your experience more personalized. 




Proactively scheduling application tasks and deadlines

Agents autonomously talk about your schedules and application materials.




Process




Week 1: User research and technical review

Insights from conversations with students and counselors 
↳ and the opportunities from AI


Tech previously couldn’t scale college consulting because it’s highly personalized. Every candidate is different.
↳ LLMs can have personalized conversations that respond to any scenario. We just have to make sure that they’re up to speed with the information of the user.

People frequently hire college counselors because they’re overwhelmed by the amount of steps in the process.
↳ LLMs are good at explaining things from a broad domain in a friendly way.

A big part of what makes a successful college counselor is time spent managing deadlines and planning. It’s more effort than tasks we’d typically expect, like finding colleges and giving general advice.
↳ LLMs take the information they get from conversations with the user, and apply these formulaic methods of deadline management.

College counseling can take anywhere between a few months to a few years.
↳ We need to find a way to extend the user journey of a chatbot from a few minutes (like how we currently use chatGPT) to the order of years.



Week 1.5: Wireframing, user testing, product ideation

I designed wireframes to make critical product decisions with my team. I conducted think-aloud interviews with 4 users with this.


I found that users are most familiar with the chatbot design pattern, so we leverage it in our designs. But we found that they aren’t able to talk with agents effectively without guidance, so we added suggestions on how to start and continue a conversation with an AI agent.

To make the experience personalized, users need to understand how we prompt chatbots to get a better sense of them. That’s why we built transparent “knowledge repositories” about what an AI agent knows about you.







Week 2: Visual design, interaction design

I created a brand identity and design system inspired by the versatility of post-it notes. 







Week 2-4: Interaction design iterations, handoff, development of demo

I worked with AI enginers to design the prompt system to ensure a good user experience.






Outcome
Demo shipped, used to raise seed capital for further development

Team learnings
Good AI systems need good context, and building a good context for AI is not just an engineering/AI problem, but also a user-facing and system-facing task. It’s important to design UX to facilitate the building of a better context.

Things I found rewarding
Visioning a product with real impact, problem-solving with a highly technical startup team with fast turnaround, effectively communicating conversation flows and AI prompting systems, successfully promoting user-centered design within the team.