QUILL

TL;DR I created QUILL (Qualitative UX Interview Logic & Learning), an interactive bot that helps UXers prep for upcoming interviews.

 

Background

Problem

As an active mentor on ADPList, I find a lot of my mentees are looking for interview prep. There can often be many different stages within a UX interview process, with many questions that UXers are expected to know. This can feel daunting, especially for those new to the field.

Insights from my mentorship sessions

  • 65% of my mentees (28 out of 43) specifically sought guidance on interviews and the job application process, ranging from strategic advice for "Big Tech" roles to technical interview practice

  • Over half of mentees (51%) requested detailed feedback on their portfolios and how to best present their work

  • Nearly 1 in 5 mentees (18.6%) requested a live "mock" practice session, indicating that simulating the pressure of a real-time portfolio run-through or behavioral round is a high-priority need

  • 25.5% of mentees (11 out of 43) asked for guidance on storytelling and presenting case studies or slide decks, emphasizing the importance of "quantifying impact" to demonstrate seniority

 

Goal

Provide UX designers the right level of feedback to help them confidently prepare for UX interviews. QUILL should be able to analyse the UXers responses and provide guidance on how to improve.

Questions UX designers have:

  • How good was their response? Would it pass in an interview setting?

  • What is the TL;DR of the feedback?

  • What worked well?

  • What needs to be improved?

  • What would a good answer look like?

 

Critical User Journey (CUJ)

As a UX designer, I want to ace my interview, so I prepare by answering practice questions and getting actionable feedback on how to improve.

 

〰️〰️〰️〰️〰️〰️〰️

〰️〰️〰️〰️〰️〰️〰️

 

Process

Collect sample questions

To ensure high quality and accurate questions, I provided my AI tooling with a list of questions that I prepared for each part of the interview process. This list is used to create a standard for newly generated questions.

Example of a standard UX interview process

 

Define evaluation criteria

UXers should be evaluated on the level of the role that they are applying to.

Example of UX Leveling

 

Crafting the initial prompt

 

Vibe coding

Screenshot from AI Studio, where I built QUILL (formerly known as QUIP)

Optimising for mobile

QUILL is build around the idea that UXers will practice with any little bit of time they have free. Because of that, the app is build with mobile in mind and only asks one question at a time.

1st iteration: Text input field and button are hidde, does not perform well for longer questions.

2nd iteration: Pulled all of the features above the fold but does not make good use of vertical and horizontal space.

3rd iteration: Clean up of redundant metadata and optimised for touch areas.

Improving waiting periods

The user is spending a lot of time waiting for the AI-generated analysis to load. To make this dead time more useful, I asked the agent to display helpful tips & tricks that are related to the UX interview process. This are stored so that they can load on demand.

Before: Simple loading icon with unhelpful text

After: UX tips & tricks are displayed and cycled to stay fresh

 

Iterating on the analysis screen

I spent many iterations trying to make the analysis screen the most valuable for users.

Applying styling

  • Providing logo / branding guide

  • Integrating dark mode

 

Debugging

Latency

To improve the perceived waiting time for users, start generating the question before the user clicks anything. This should be done on the backend so that when the user does click the button, the question will load very quickly.

Gemini API issues

I ran into a lot of issues with loading the AI analysis and feedback. This ended up being due to an issue with the connection to the Gemini API. I resolved it by asking AI Studio to remove the touchpoint and then re-add it using engineering best practices.

 

Getting Feedback

Heuristic Evaluation

I attached a demo video of my prototype to an AI agent and then provided a prompt that analyses against Jacob Nielsen’s 10 Heuristic Principles and provides guidance on how to improve. Here is the guidance I received:

  • Overall Rating: High Usability The app is highly intuitive and focused. It successfully bridges the gap between a "mock interview" and "actionable coaching."

  • Heuristics to improve

    • User Control and Freedom: Users can navigate back via the "Change Settings" link if they want to restart. A more flexible system would allow an "on-the-fly" adjustment (e.g., a small dropdown or toggle on the question page).

    • Consistency and Standards: Duplication of metadata. Usually, the top right is reserved for global status or account info, while the area above the main content is for "contextual" info. Using both makes the user's eye jump back and forth, wondering if there is a subtle difference between the two labels that they might be missing.

    • Aesthetic and Minimalist Design: Utilize vertical and horizontal space more efficiently to showcase the content better.

 

Outcome

 

Next steps

  1. Making the voice experience less jarring

  2. Adjusting the styling & dark mode

  3. Testing with UX designers

 

Learnings

Coming soon… Still collecting my thoughts…

Previous
Previous

Grape Job

Next
Next

Eliza-Wrapped 2025