Designing a clearer, more context-aware AI career coach for students
Problem
Students weren't completing the assessment because the AI felt repetitive, confusing, and untrustworthy.
Results
20% increase in task completion and 10% faster completion with voice interactions
Timeline
6 months
My Role
Product Design Lead, UX Research
Team
2 Product Designers
2 Software Engineers
The original assessment took 1 hour to complete. After switching to a faster LLM, we reduced completion time to 20–30 minutes. We tested the experience with 10 students and worked with a 6-person advisory board of vocational psychologists. We found that students were struggling to understand and trust the AI.
Because students and the advisors did not trust in the efficacy of the AI prompts and questions, I defined 3 principles for a trustworthy AI experience:
Be clear. Ask one question at a time using language students naturally understand
Be aware. Carry context forward so students feel heard
Be consistent. Create repeatable patterns for how the AI communicates
These principles are outlined in a pattern library that defines how the chatbot should ask questions, acknowledge responses, handle uncertainty, and recover from errors.
An AI Pattern Library to define how the chatbot should interact with the user
For students already comfortable speaking with technology, voice created an opportunity to reduce friction.
I designed a multimodal experience supporting voice-only and voice-to-text interactions, allowing for a more natural and accessible experience.
Before
After
increase in task completion
faster completion time with voice interactions
The MVP is preparing for a pilot with First Gen Scholars, a nonprofit supporting first-generation and low-income students pursuing college opportunities. Hundreds of students from this community will be among the first to use the redesigned experience.
Takeaway
The most effective decisions came from listening to users and stakeholders to define guardrails for safety and effectiveness for each assessment.
