01
Retry behavior
Test whether users speak more clearly after one focused takeaway.
A native iOS interview coach that turns spoken answers into context-aware feedback for the next attempt.
[Role]
Product Designer / Prototyper · Team Project
[Duration]
July 2026 — Present
[Tech Stack]
SwiftUI, Apple Speech, OpenAI API, Figma
AI UX
Conversational UX
Voice Interaction
01 / The problem
Interview feedback often sounds useful without actually knowing the candidate.
Target User: Interview Coach is designed for early-career candidates stepping into new roles who need to practice not only what to say, but how to connect their experience to what a specific role is looking for.
The problem: Most interview practice tools can tell you what you said.
They can transcribe an answer, identify structure, and generate suggestions such as:
“Add more detail.”
“Quantify your impact.”
“Use the STAR method.”
But those recommendations often miss a more important question:
Is this a strong answer for this person, this role, and this interview?
Your actual answer
Generic feedback
02 / The opportunity
Good feedback needs context.
The quality of an interview answer depends on more than the transcript.
To understand what matters, the coach needs to know four things.
Instead of judging an answer in isolation, Savvy uses the candidate’s experience, the role, and the spoken response to decide what feedback matters most.
03 / Design Principle 1
Context before coaching.
Generic interview practice starts with a question. Savvy starts by understanding what the candidate is preparing for. Before practice begins, the experience builds a shared context from the role, the coaching style, and the candidate’s own experience.
01 Job description
Understand what the role is looking for
02 Coaching style
Decide how the practice should feel
03 Resume
Ground coaching in real experience
04 Question
Turn context into relevant practice
04 / Design Principle 2
Feedback only matters if it changes the next answer.
Instead of treating feedback as the end of the experience, I designed Interview Coach around a repeated learning loop.
01 Prompt
Role-specific prompt from the job post
02 Answer
Speak naturally and see a live transcript
03 Feedback
See one prioritized takeaway with evidence
04 Retry (2nd attempt)
Try the next version right away
05 / Speak
Keeping the interface out of the conversation.
Early concepts exposed the answer structure too aggressively. The framework was useful, but it competed with the question and made the experience feel like filling out a form.
I moved guidance into a secondary layer, kept the question dominant, and let the transcript stay low in the hierarchy while the user speaks.
Before
Structure dominated the experience.
After
The question leads; guidance supports when needed.
Design decision
Keep coaching available without making the user perform for the interface.
06 / Review
Turn feedback into a next action.
Early feedback screens gave users an improved answer, but they still had to figure out what mattered most. I redesigned the review state to lead with one prioritized takeaway, show the evidence behind it, and connect that insight directly to retry.
Before
The product gave users a better answer to try.
After
The product gives users a reason to retry.
Design decision
Turn feedback into a next action, not just a better answer.
07 / Trust
A clearer answer should still sound like the same candidate.
The improved answer is a practice target, not a replacement. Savvy can restructure, shorten, clarify, and correct what the user said but it should not invent experience, outcomes, numbers, or claims.
Before
What the candidate said
“I worked on an interaction prototype at Designworks and helped explore different concepts.”
Original input from the candidate.
Allowed
Clarify what was already there
“At Designworks, I explored multiple interaction concepts through rapid prototyping.”
Improves clarity without changing the facts.
Not allowed
Do not invent evidence
“At Designworks, I led a team that increased engagement by 30%.”
Adds unsupported claims or fabricated evidence.
Design rule: improve clarity without changing the facts.
08 / Coach behavior
Coach style changes behavior, not just tone.
I stopped treating AI personality as tone. Each mode changes the amount of guidance, the directness of follow-up questions, and the intensity of the rewrite.
Before
The first version treated coaching style as personality controls.
After
I shifted from abstract sliders to explicit coaching modes, so users could understand what would actually change during practice.
Design decision
Replace abstract personality tuning with clearer, behavior-based coaching modes.
09 / Prototype
I needed to test the conversation, not just the screens.
A static prototype could show the flow, but it couldn't answer the questions I cared about:
Does live transcription distract from speaking?
How quickly should feedback appear after an answer?
Does one takeaway actually encourage a retry?
Does the second attempt feel like practice rather than correction?
So I built the interaction in SwiftUI using Apple Speech and the OpenAI API, allowing me to test the complete loop with real spoken input.
01
Voice
Spoken answer
→
02
Apple Speech
Live transcription
→
03
Transcript
What the user said
→
04
Context + OpenAI
Role-aware analysis
→
05
Feedback
One focused takeaway
→
06
Retry
Second attempt
Design decision
Build the prototype close enough to the real interaction so the spoken practice loop could be tested, not just imagined.
10 / Next
Testing whether feedback actually improves the next attempt.
The next question is not whether the AI can produce a better answer.
It is whether the feedback helps users give a clearer second attempt.
02
Follow-up depth
Explore when Savvy should ask a follow-up instead of rewriting.
03
Progress over time
Track recurring patterns across clarity, structure, role fit, and impact.
What I learned
I learned that the value of AI coaching isn't in producing the perfect answer. It's in giving the user just enough clarity to make their next answer better.
All Works