top of page
BOLD: Build a resume with AI
PRODUCT DESIGN l USER RESEARCH l A/B TESTING
INTRODUCTION
BOLD is an online platform that offers a tools to help job seekers to build their resume, cover letter and look for jobs. I worked on the Resume Builder team and led the end-to-end design of the GPT-first mobile resume builder, a new feature for the mobile web app. This AI-powered tool enables users to quickly generate personalized resumes aligned with their work experience and skill set. The feature is set to launch in end of August 2025.
MY ROLE
Led end-to-end design process from initial discovery, sketch, wireframes, user testing, Hi-Fi prototype, and final ideation.
MY TEAM
UX Designer (Me), 1 Design Director,
1 Product Manager, 10+ developers,
1 UX Writer, 1 UX Researcher,
2 AI Content strategist
TIMELINE
May 2025 - Aug 2025
USER PROBLEM
Why do we need this new feature?
We discovered that mobile users had a significantly higher bounce rate of 55% compared to 35% on desktop. Mobile users often have shorter attention spans and face more frequent distractions, which means they expect fast, intuitive, and mobile-native experiences. This insight highlighted a critical need to optimize the resume-building flow specifically for mobile.
What does the current mobile resume-building process look like?
Get started
Answer onboarding questions and pick a template
Build a resume
Go through 6 sections and
fill in the details
Finalize & Download
View the resume and do
the final editing
BREAKDOWN OF THE PROBLEM
Lengthy multi-step process
Users have to navigate through multiple steps and pages to add each section of their resume. This repetitive and fragmented experience creates friction and increases frustration.
.png){kind=small}
Keyboard heavy input
The current mobile resume building process mirrors desktop interactions relying heavily on keyboard input, but smaller screens make tying and editing cumbersome.
Lack of automated tools
The experience lacks automation or intelligent assistance, requiring users to manually input all resume content. This increases cognitive load and makes the process feel time-consuming and exhausting.
HOW MIGHT WE
Create a mobile resume-building experience that feels more intuitive, automated, and aligned with mobile-first behaviors?
OUR SOLUTION
GPT Resume Builder
To reduce friction and simplify the mobile resume-building process, we introduced a new feature that enables users to engage in a brief back-and-forth conversation with GPT. This experience combines a short chat, guided prompts, and a structured webform that feeds into GPT, resulting in a personalized draft resume. It offers users a simple, mobile-friendly way to kickstart their resume with less effort and more confidence.
GOAL
What are we trying to achieve in MVP?
Since this new AI feature differs significantly from our existing resume-building flow, we decided to test MVP concepts to help us validate whether there’s a user segment interested in an AI-driven experience, not to drive immediate conversions.
We aimed to gather valuable insights from this test to inform future development and ultimately launch a refined, user-ready feature by the end of the year.
To do so, we focused on three key areas:
Mobile first
Design a quick, intuitive experience tailored to mobile users, prioritizing speed and simplicity to feel natural and frictionless on smaller screens.
AI integration
Integrate conversational AI to guide users through brief prompts, reducing cognitive load and making resume creation more approachable.
Voice to text
Introduce voice input as an alternative to typing, reducing friction and making the process faster, more accessible, and better suited to mobile patterns.
USER RESEARCH
Second rounds of voice prompt testing
First, we conducted two rounds of user testing via UserTesting.com to evaluate how different voice prompt strategies impact the quality of spoken input for GPT-based resume generation.
#1st round: Understanding users response (16 participants)
The first round focused on identifying prompt wordings that encourage clear, relevant, resume-style responses. We tested two versions of prompt: bullet-point style and paragraph style.
What users said
“I probably left out a lot. It was hard to remember everything at once.”
.png)
“I don’t know if I’m saying enough. An example would really help here.”
“The bullet-point style prompt is easier to follow and feels more guided. ”
Insights
✅️ The bullet points perform better than the paragraph.
⚠️️ 4/16 people requested to see an example of what a good response is like.
⚠️️ 6/16 people requested follow-up questions.
❌️ In the paragraph prompt, 4/16 people provided too little or vague answer, and wanted to see more target questions to think through answer through instead of doing it on the spot.
#2nd round: Testing prompts flow (14 participants)
The second round evaluated which prompt flow produced the highest-quality input: the single-prompt flow, where users described their entire work history in one recording. The multi-prompt flow, where each work history entry was recorded separately, and the multi-prompt + skills flow, where users recorded each work history entry along with a separate voice prompt for skills.
What users said
“I wish I could talk about each job separately.”
.png)
“I said things like ‘team player’ because I wasn’t sure what else to add.”
.png)
“Multi-prompt flow feels more manageable. I know exactly what to say.”
.png)
Insights
✅️ Multi-prompt flow yielded high-quality input. Users consistently provided the job title, companies, dates, and multi-responsibilities. They closely followed the prompt format.
⚠️️ Skills responses were vague. User defaulted to soft skills or generic terms.
❌️ Single-prompt flow performed poorly since it was overwhelming and disorganized. They often skipped over details beyond the job title and company.
USER FLOW
Three different user flows
Based on our research insights, we removed the single-prompt flow, kept the multi-prompt flow, and introduced a new version with breakdown questions for a more guided experience. We also added a skills section to two flows, giving users the option to include more detail about their strengths.
DESIGN DECISION
Create an AI-powered resume in minutes
Using user insights and flow analysis, we identified and prioritized the core features that would shape the AI resume-building experience for our initial test.
Option to speak or type
We introduced a toggle option that allows users to input their answers via either voice or text. This flexibility empowers users to choose the method that feels most natural and convenient, especially important for mobile users who may prefer speaking over typing.
Guided prompts & examples
To ensure reliable and high-quality resume generation, we designed clear, structured prompts accompanied by contextual examples. This guidance helps users understand what to say or write, reducing confusion and increasing the chances of getting relevant, resume-style output from GPT.
Error handling & fallbacks
Because AI can be unpredictable, we prioritized designing helpful error states. These include reassuring messages, fallback options, and clear next steps so users never feel stuck, misled, or out of control. Our goal was to maintain trust and flow, even when the system doesn’t respond perfectly.
Hi-Fi DESIGN
Final features
Create with AI
We introduced a new “Create with AI” button, positioned below the existing options (Start from Scratch & Upload) as an additional entry point for building a resume. The first step in this flow is the AI landing page, which helps users understand the feature and prepare before getting started. Next, users see the contact information page. Since contact details are essential for any resume, we included this step upfront to give users the option to fill it in as part of the flow.
Speak work history
In the AI flow, the Speak page appears by default, as most users prefer speaking over writing. They can switch options anytime via the top tab menu. Here, users describe their work history by following a prompt and tapping the speaker button, which activates listening mode with a sound wave animation. Entries can be deleted with Discard or submitted with Done. Once enough information is collected, GPT takes users to the overview page to review job titles and employers, with an illustration adding a touch of delight on the first job page.
Write work history
If users prefer typing, they can switch to the Write tab. This flow includes fields for job title, employer, and a manual input field for job descriptions to ensure accurate details. From the second job onward, users see a list of all added jobs, with a trash icon to remove any incorrect entries recorded by GPT.
Example script
In both the Speak and Write flows, users can tap the See Example button to view a sample script. It’s hidden by default to prevent distraction while speaking into the microphone.
Skills section
While adding skills can help GPT create a stronger resume, users can choose to skip this step. We provide pre-generated suggestions based on all entered jobs to make it easy for users to pick the ones they want.
Follow-up questions (Generated by GPT)
We set to ask one mandatory follow-up question to capture any missing details for a stronger resume. If essential information is still missing, GPT will ask up to two additional follow-up questions before taking users to the overview page.
Resume Finalize page
This is page where users can see the final resume generated by GPT. We added the tooltip pointing the “Update content” button where users can add/edit their content. We plan to iterate Finalize page in V2 to highlight sections.
KPI
Key metrics to track
These are the key metrics we are hoping to gain from the initial A/B testing.
Post-creation editing
Average number of edits per section after GPT resume generation compared to the current baseline.
Prompt effectiveness
Percentage of users receiving a high-quality resume (score >6/10) on the first GPT output vs. needing a second attempt.
Input method preference
Percentage of users who choose voice vs. text for initial input.
Builder completion speed
Average time to complete the resume builder using the AI feature vs. the standard flow.
Segment interest
Percentage of users who engaged with the AI-only resume builder.
Drop-off rates
Comparison of user drop-off rates on the AI flow vs. the baseline flow.
NEXT STEP
What’s next?
-
Await A/B testing results to gather meaningful insights on how users interact with the voice input feature, and explore further development if validated.
-
Enhance the resume editing experience, allowing users to refine AI-generated content into a more tailored, polished resume.
-
Assess the accuracy of GPT-generated resumes to determine if the current prompting strategy needs refinement for more consistent, high-quality outputs.
LEARNINGS AND TAKEAWAY
What was challenging and what went well?
Technical Challenges & Timeline Constraints
This project was complex, with multiple user flows technical constraints, and a tight timeline. Microphone issues extended UAT by over a month, delaying launch. Despite the challenges, I worked closely with the engineering team in India to troubleshoot, address design issues, and deliver clear handoffs with animated prototypes, ensuring alignment and high-quality implementation.
Cross-Team Collaboration & Iteration
This project involved a wide range of stakeholders across Product, AI, Content, and Design. Collaborating closely with these teams taught me the importance of strong communication to align design decisions with business goals and deliver a cohesive user experience. Throughout development, we went through continuous iterative process to validate GPT prompts and reviewed design features. Each iteration brought us closer to delivering seamless and user-friendly resume-building experience.
bottom of page