JobNova AI Job Search Platform

Designing a 0-to-1 AI job-search system that helps users complete relevant applications faster without giving up trust or control.

0-to-1 AI SaaSProduct StrategyUX ResearchAI WorkflowDesign System

About JobNova

Most AI job-search products automate individual tasks. JobNova redesigns the entire application workflow, from discovering the right opportunities to completing trustworthy applications.

View live site

Role

Founding Designer

End to end product design from research to delivery

Team

2 Designers (Me & 1 Intern)

1 Product Manager (Founder)

1 Full-Stack Engineer

2 ML Engineers

Tool

Company

Nova AI

Year

2025 - Ongoing

01 / The fragmented journey

AI tools optimized job-search tasks.
The journey stayed fragmented.

Existing tools helped users write, score, autofill, or track pieces of the process. But users still had to stitch together discovery, evaluation, application, and follow-up across disconnected products.

The product advantage was not another isolated AI feature. It was the orchestration layer connecting discovery, judgment, content, execution, and follow-up.

02 / How I worked

The work ran as a loop, not a handoff.
Discover, reframe with the team, then review what shipped.

Early Opportunity

Initial concept
Fragmented workflow identified
Early product hypothesis defined

CROSS-TEAM TOUCHPOINTS

Team

Aligned research goals and product priorities

Discovery & Scoping

20+ user interviews and competitive research
Defined the core job-search problems and opportunity areas

CROSS-TEAM TOUCHPOINTS

Product

Refined the problem space and feature priorities

Checked whether AI could support the proposed concepts

Reviewed platform and integration constraints

Define & Design

Reframed the product from auto-apply to user-governed automation
Designed the core flow: Match → Customize → Apply → Track
Created wireframes, prototypes, and high-fidelity designs

CROSS-TEAM TOUCHPOINTS

Product

Reviewed scope and design decisions at each iteration

Explored feasibility for matching, resume generation, and automation

Reviewed implementation feasibility before handoff

Build

QA & Review

Reviewed live flows for design consistency
Checked errors, empty states, and recovery paths

CROSS-TEAM TOUCHPOINTS

Direct back-and-forth to resolve UI and flow gaps

Reviewed output quality and failure cases

Launch

Post-Launch

Reviewed product data and user behavior
Identified friction and fed insights into the next cycle

CROSS-TEAM TOUCHPOINTS

Product

Shared findings to inform roadmap priorities

Reviewed model performance and failure patterns

Investigated product and integration issues

Early Opportunity

Initial concept
Fragmented workflow identified
Early product hypothesis defined

CROSS-TEAM TOUCHPOINTS

Team

Aligned research goals and product priorities

Discovery & Scoping

20+ user interviews and competitive research
Defined the core job-search problems and opportunity areas

CROSS-TEAM TOUCHPOINTS

Product

Refined the problem space and feature priorities

Checked whether AI could support the proposed concepts

Reviewed platform and integration constraints

Define & Design

Reframed the product from auto-apply to user-governed automation
Designed the core flow: Match → Customize → Apply → Track
Created wireframes, prototypes, and high-fidelity designs

CROSS-TEAM TOUCHPOINTS

Product

Reviewed scope and design decisions at each iteration

Explored feasibility for matching, resume generation, and automation

Reviewed implementation feasibility before handoff

Build

QA & Review

Reviewed live flows for design consistency
Checked errors, empty states, and recovery paths

CROSS-TEAM TOUCHPOINTS

Direct back-and-forth to resolve UI and flow gaps

Reviewed output quality and failure cases

Launch

Post-Launch

Reviewed product data and user behavior
Identified friction and fed insights into the next cycle

CROSS-TEAM TOUCHPOINTS

Product

Shared findings to inform roadmap priorities

Reviewed model performance and failure patterns

Investigated product and integration issues

03 / The first hypothesis

We assumed AI could replace pieces of the job-search journey.

The first bet was a stack of AI features: find matching jobs and explain why they matched, generate application materials, then add job training and referrals from our own resources.

Job matching

Job details

Version 1 job matching: match score, explanation, and generate-resume action on a job detail page.

Tailored resume

Outcome

AI made parts of applying faster.

Users could see matched roles from multiple platforms in one place, generate tailored materials for a specific job, and find people to contact. AI raised the efficiency of those steps. They no longer had to hop between boards just to start.

The problem

The journey still ran on manual work.

Users still had to judge which roles were worth applying to, submit each application themselves, and track every outcome one by one. Faster pieces did not remove the labor of the full process.

04 / What research showed

Matching and writing were not the bottleneck.
Users needed applying to happen without them.

Across 20+ interviews, faster matching and tailored materials still left the same labor in the user's hands: decide, submit, and track, one role at a time.

“By the time I tailor everything, the best opportunities already feel out of reach.”
Career transitioner, research interview

The product direction

Automate the application. Users should only need to focus on interviews.

05 / One complete application journey

Users would focus on interviews.
The product had to apply.

That meant one complete application journey: match, customize, apply, and track. Auto Apply became the product that ran that loop, so users were no longer judging, submitting, and following up one role at a time.

06 / Four levels of AI responsibility

The challenge was not whether to automate applying.
It was how to balance autonomy and control.

Auto Apply would let users leave the forms. It would also send applications in their name. The product had to hold both at once, at every step of the loop.

Autonomy

If users still judge, submit, and track one role at a time, Auto Apply is not doing the job. The system has to take the repetitive work so they can leave the forms.

Control

Applying on someone’s behalf is high-stakes. A submitted application cannot be taken back. Users need to set where AI may act before it sends, see what went out, and stop or tighten the rules so it does not keep going.

That balance needed four moments of AI responsibility, not four separate features.

Match, customize, apply, and track had to work as one loop. Each step changed how much the AI was allowed to do, and none of them could succeed alone.

07 / A permission system, not a switch

Applying on someone’s behalf is the highest-stakes moment in the loop.

Applying on someone’s behalf is the highest-stakes step. AI only acts inside boundaries the user has already set.

The first spec was an on/off switch. It broke once login expiry, queues, and retries needed named states. We kept the switch and wrapped it in matching strategy, autonomy, materials, and notifications.

Before

Users define where AI may act: match threshold, job preferences, resume choice, and notification rules.

During

AI continues only when conditions are satisfied; exceptions require approval.

After

Every action remains visible and recoverable.

Annotated JobNova controlled Auto Apply settings design

Applied4 hours ago112 applicants

84%

Good Match

H1B Sponsor Likely

User Experience Designer

Cursor AIAutomotive • Big Data • Growth Stage

Nashville, TN

Full time

Remote

Mid Level

$112K/yr - $152K/yr

Your application was automatically submitted.

Applied4 hours ago112 applicants

84%

Good Match

H1B Sponsor Likely

User Experience Designer

Cursor AIAutomotive • Big Data • Growth Stage

Nashville, TN

Full time

Remote

Mid Level

$112K/yr - $152K/yr

Your application was automatically submitted.

Trust is earned

We wanted full autonomy. Users never want to be out of control.

We gave transparency and control first, so trust could build. Autonomy is what users grant after that, not something they hand over up front.

Day one

37%

turned on Apply without approval from the start

Within a week

82%

had moved to Apply without approval

08 / Every action leaves a trace

If an application disappears after send, the automation has failed.

The loop is not complete at submission. Users still need to see what happened, what is happening, and what needs their attention next.

History

What happened and what materials were submitted.

Status

What is happening now across applications and recruiter responses.

Attention

What requires user review, correction, or follow-up next.

Annotated JobNova application tracking overview

Annotated JobNova application and inbox tracking design

09 / From screens to product rules

The four moments had to feel like one product.

I translated the trust model into reusable states, components, and interaction rules so Match, Customize, Apply, and Track behaved consistently.

JobNova design system covering colors, typography, spacing, components, icons, cards, filters, and application states.

10 / Results, debt, and learnings

The goal was never maximum automation.
It was automation people could understand, configure, and hold accountable.

Launch numbers show activation and execution. They do not, by themselves, prove that users trusted Auto Apply.

Activation

68.06%

30-day onboarding completion

Adoption

19.39%

30-day Auto Apply activation

Reliability

88.21%

Final application success rate

Speed

1.03 hrs

Match-to-application time

Outcome

10.77%

Interview invitation rate

What experience can still raise

Submission success and interview rate sit downstream. These two numbers can still move by making setup easier to finish.

30-day onboarding completion

68.06% finished resume upload and job preferences (179 of 263). That is a setup experience, not an execution metric. Shortening and clarifying onboarding is how this number moves.

30-day Auto Apply activation

19.39% enabled Auto Apply (51 of 263). Among those who finished onboarding, 28.5%. Auto Apply setup is still too long. Reducing that friction is how this number moves.

Three learnings I will carry forward

Explain before acting

Trust starts with understanding.

Make control configurable

Users decide where AI can act.

Design beyond execution

Automation needs history, recovery, and accountability.

1. The Product Thesis

Most AI job-search products automate individual tasks. JobNova redesigns the entire application workflow—from discovering the right opportunities to completing trustworthy applications.

Match, resume customization, Auto Apply, and tracking work as one system—not four disconnected features.

View live site ↗

Workflow AI,
not Feature AI.

Designing trustworthy AI wasn’t about adding more automation. It was about connecting every step of the application journey while keeping decisions visible and controllable.

More AI tools.
Still a fragmented application workflow.

Users switched between job boards, resume editors, ATS checkers, AI writing tools, email, and spreadsheets just to complete a single application.

The product advantage was not another isolated AI feature. It was the orchestration layer connecting discovery, judgment, content, execution, and follow-up.

Early evidence

Product adoption and completed actions

1,240

Registered users

68.06%

30-day onboarding completion

54%

Used AI resume customization

19.39%

Enabled Auto Apply

8,700+

Applications completed

2. Why Another AI Job Search Product?

Existing AI products optimized individual tasks.
The opportunity was redesigning the entire application journey.

Most AI job-search products focused on solving one step of the process—writing resumes, autofilling applications, checking ATS compatibility, or tracking submissions.

While these tools improved individual tasks, users still had to move between multiple platforms to complete a single job application. The workflow remained fragmented.

Existing Landscape

A category organized around isolated tasks

Each transition moved context, decisions, and materials into another product.

Discover Jobs

LinkedInIndeed

Resume & ATS

TealReziKickresume

Applications

SimplifyLoopCV

Tracking

HuntrSpreadsheet

Users stitched together multiple AI tools to complete one application.

JobNova

One product.
One continuous workflow.

Match

Customize

Apply

Track

Design Opportunity

Instead of building another AI resume assistant, we reframed the problem:

How might we help users complete an entire application journey with AI while keeping them in control of every important decision?

Key Product Principles

Three rules guided every product decision

Explain before automating

Help users understand recommendations before AI takes action.

Keep users in control

Users define when AI can act and when approval is required.

Connect the entire workflow

Reduce context switching by bringing discovery, customization, application, and tracking into one experience.

5. Trust by Design

We tested where the process broke, not which features people wanted.

Before defining the product, we needed to validate whether efficiency was truly the biggest blocker and how much control users were willing to hand over to AI.

Research Goal

We wanted to understand where the job-search process actually broke down, how people judged whether a role deserved effort, and which decisions they were willing to delegate to AI.

Where do users lose time or abandon an application?

How do users decide whether a role is worth applying to?

Which tasks are safe to delegate to AI?

Which decisions must stay under user confirmation?

Discover

Evaluate

Prepare

Apply

Track

Who we're designing for

Alex Chen

Entry-Level Software Engineer (Job Seeking)

1st-Time Job Seeker

Applying everywhere, hearing back from no one

I’m applying every day, but I have no idea why nothing is getting through.

Recent graduate seeking his first full-time role, struggling to judge fit, understand ATS rejection, and prioritize the right opportunities.

Sarah Müller

Senior Marketing Manager

Career transitioner

Rewriting every resume, missing the right moment

By the time I tailor everything, the best opportunities already feel out of reach.

Experienced professional transitioning into product management, balancing application quality with limited time outside her current job.

What We Focused On

Transparency

The AI is doing complex work behind the scenes. How much of that process should users see — and in what form?

Human Control

Where does autonomy end and overreach begin?

Explainability

Results mean nothing if users can't evaluate them. How do you surface reasoning?

Credibility

AI can be wrong. How do you present output users feel equipped to trust?

These feel like separate problems. But the more we dug in, the more they all pointed to the same thing.

Not about transparency, not about control. It's a trust problem.

Method

20+

semi-structured interviews across real job-search journeys.

First-time job seekersEmployed job seekersCareer transition users

Interview coverage

How users search, filter, and judge roles
How much time resume tailoring and materials take
Why users delay or abandon applications
How users track application status
Where users draw trust boundaries around AI

Insight 01

Relevance mattered more than volume

Users already had access to many roles, but struggled to decide which ones deserved the effort of tailoring and applying.

Matching needed explanation, not just more listings.

Insight 02

Opaque AI created new review work

When users could not see why a role was recommended or what changed in a resume, they checked AI work manually.

Automation had to reduce work without hiding decisions.

Insight 03

Control needs varied by user

First-time job seekers wanted more guidance. Experienced users wanted precision, personal expression, and application quality.

The product needed levels of control, not one automation switch.

Insight 04

Submitting was not the end

After applying, users still needed to confirm submission, review the material version, track replies, and adjust strategy.

The loop had to extend from Apply to Track.

Reframing the Assumptions

Research did not reject automation. It changed what successful automation needed to mean.

Initial assumption

What research showed

Efficiency is the biggest barrier

Partly true. Users also struggled with screening difficulty and opaque feedback.

More applications create more opportunities

Not enough. Users cared more about relevance and meaningful replies.

Users will let AI apply automatically

Conditionally true. Users needed explanations and execution boundaries.

Discover, Customize, Apply completes the loop

Incomplete. Users still needed tracking and feedback management.

Research changed the product from an automation tool into a controlled AI system.

From

Help users automatically complete more applications.

Help users complete more relevant applications faster, while keeping control of key career decisions.

How might we help job seekers complete more relevant applications faster, without losing trust and control over AI-driven decisions?

7. Design Decisions

Four moments where trust had to be designed into the workflow.

The strongest design decisions were not isolated screens. They were interaction rules that clarified what AI knew, what it changed, and when it was allowed to act.

Design Decision 01

Explainable Job Matching

Making AI recommendations understandable

How can AI help users make confident decisions, instead of simply providing more recommendations?

1. The Design Challenge

Users received a high volume of job recommendations every day. The difficult part was not finding another role—it was deciding which opportunity deserved the time required to apply.

Most AI products returned a Match Score without explaining its basis, forcing users to reread the job description and verify the recommendation themselves.

Users didn’t lack recommendations—they lacked confidence in acting on them.

2. What We Learned

A higher Match Score did not automatically make a recommendation more believable.

Recommendations without explanation still created manual verification work.

Users wanted help understanding the opportunity—not a system that made the decision for them.

Trust comes from understanding, not prediction accuracy alone.

3. Design Decision

We shifted from prediction to explanation.

Improving the score alone would make the system appear more precise without reducing the user’s verification burden. We treated explainability as the primary interaction problem: surface the evidence behind the recommendation, then let the user decide.

Before

After

AI gives a score

AI explains the score

Recommendation first

Evidence first

System decides

User decides with evidence

Only showed a combined match percentage.

Killed

Put the full rationale on every job card. Scanning collapsed.

Shipped

Kept the score on the list; moved evidence into details.

4. Design Principles

Three principles translated explainability into interaction rules.

Prioritize before explaining

The list helps users scan and rank opportunities before investing attention.

Explain before asking users to act

The detail view provides the recommendation evidence, not only a score.

Highlight uncertainty

Skill gaps remain visible instead of presenting AI judgment as complete certainty.

5. Translating the Decision into the Experience

The interface made evidence available at the moment each decision required it.

JobNova job list showing match score, key role information, salary, work mode, and saved state.

JobNova job detail experience explaining fit, skills, role requirements, and resume suggestions.

C. Skill Gap

Make uncertainty visible

Showing what the AI didn’t know increased confidence more than hiding uncertainty.

Matching Skills

Interaction design

User research

Prototyping

Cross-functional work

Missing Skills

Enterprise systems

Design leadership

Domain-specific tooling

6. Outcome

Feature-level signals, not whole-product metrics

Match explanation view rateJob detail to application conversionApplication rate by match bandSkip or “not interested” rateDecision time before and after viewing explanations

7. Key Takeaway
Helping users understand a recommendation was ultimately more valuable than making the recommendation appear more intelligent.

Design Decision 02

AI Resume Customization

Preserving user authorship in AI-assisted editing

How can AI improve application materials while ensuring users remain the author of their own experience?

1. The Design Challenge

Tailoring a resume for every application was repetitive and time-consuming. Users wanted AI to accelerate the process, but they were uncomfortable submitting content they had not reviewed.

Blindly accepting a generated resume reduced confidence and raised concerns about accuracy, exaggeration, and whether the result still represented the user.

Users wanted AI to speed up writing—not to speak on their behalf.

2. What We Learned

The core concern was not whether AI could write. Users needed to know whether the result still reflected their experience, exactly what had changed, and whether each edit could be trusted.

Is this still my experience?What exactly changed?Can I trust these edits?

Transparency mattered more than automation during content creation.

3. Design Decision

We designed AI as a collaborator, not the final author.

Instead of replacing the document with a finished AI output, the system improved the user’s existing resume through visible suggestions and explicit approval.

Before

After

Generate a new resume

Improve the existing resume

Replace user content

Suggest visible changes

AI owns the draft

User approves every change

4.1

Preserve authorship

AI can suggest improvements, but the final professional story belongs to the user.

4.2

Make every change visible

Edits are highlighted and explained instead of silently replacing original content.

4.3

Review before submission

AI-generated changes require user confirmation before they become application material.

5. Translating the Decision into the Experience

Three mechanisms preserved ownership without slowing the user down.

A. Overview Before Editing

Understand the scope first

Resume scoreSuggested improvementsSection summary

Users first understand the scope of AI suggestions before entering detailed editing.

B. Transparent Editing

Keep every modification inspectable

Highlighted changesAccept / RejectInline editing

Every AI modification remained visible and editable.

C. User Approval

Let the user make the final decision

Final previewSaveGenerate PDFContinue to Apply

AI prepared the content, but users made the final decision.

JobNova resume overview and editor showing visible AI changes, section controls, and final document preview.

6. Outcome

Resume-specific validation signals

AI resume completion rateDirect-use rate after generationEditor entry rateManual edit or restore rateTime from generation to applicationGeneration failure and abandonment rate

7. Key Takeaway
AI became a writing partner rather than an automatic author, allowing users to move faster without giving up ownership of their professional story.

Design Decision 03

Controlled Auto Apply

Defining the boundaries of AI autonomy

How much control should users give AI when applying for jobs?

1. The Design Challenge

Applying for jobs is repetitive, making it an ideal candidate for automation. But submitting an application is also a personal career decision. Fully autonomous AI could save time while introducing uncertainty and unwanted actions.

The challenge wasn’t whether AI could apply automatically, but whether users would trust it to do so.

2. What We Learned

Users had fundamentally different tolerance for AI action. Some wanted full automation, some needed approval before every application, and others only wanted AI to prepare materials.

Full automationApprove every applicationPrepare materials only

Users didn’t want the same level of automation—they wanted different levels of control.

3. Design Decision

We shifted from automation to configurable autonomy.

Instead of deciding how much AI should automate, we let users define the conditions, permissions, and review boundaries under which AI could act.

Before

After

One-click Auto Apply

Rule-based automation

Same behavior for everyone

User-defined autonomy

Automation as a feature

Automation as a configurable system

Auto Apply was designed as on or off.

Broke

Shipped

Kept the switch, wrapped it in matching strategy, autonomy, materials, and notifications.

4.1

AI acts only within explicit user rules

The system cannot silently expand the role, material, or action permissions a user granted.

4.2

Configure autonomy before automation begins

Control is established before execution rather than added as a recovery mechanism afterward.

4.3

Every automated action remains visible

Queued, processing, approval, submission, and failure states stay inspectable.

4.4

Recovery is part of automation

Failures expose their cause and preserve a clear retry or manual continuation path.

5. Translating the Decision into the Experience

AI earns the right to act by following the boundaries a user defines.

A. Define the Rules

Permission before action

Users first define the conditions under which AI is allowed to act.

Match Threshold

Define the minimum role relevance AI may act on.

Job Preferences

Limit execution to the user’s role, location, salary, and work-mode rules.

Resume Selection

Choose which base material AI may tailor and submit.

Notification Preference

Decide which actions and outcomes require attention.

B. Execute Within Boundaries

Visible system action

AI executes only when predefined conditions are satisfied. Approval, queue position, processing, submission, and failure remain visible.

Approval Required4 hours ago112 applicants

84%

Good Match

H1B Sponsor Likely

User Experience Designer

Cursor AIAutomotive • Big Data • Growth Stage

Nashville, TN

Full time

Remote

Mid Level

$112K/yr - $152K/yr

Your explicit approval is required — will not auto-submitApprove now to apply

This application won't be submitted until you explicitly approve it.

Approval Required4 hours ago112 applicants

84%

Good Match

H1B Sponsor Likely

User Experience Designer

Cursor AIAutomotive • Big Data • Growth Stage

Nashville, TN

Full time

Remote

Mid Level

$112K/yr - $152K/yr

Your explicit approval is required — will not auto-submitApprove now to apply

This application won't be submitted until you explicitly approve it.

C. Handle Exceptions Gracefully

Recovery is part of the flow

When automation could not continue, the system explained why and offered a clear recovery path.

Platform unsupported

E01

Keep the match and direct the user to a manual application path.

Missing information

E02

Pause execution and ask only for the information required to continue.

Login expired

E03

Explain that authentication stopped the flow and preserve the prepared application.

Submission failed

E04

Retry safely, expose the failure reason, or hand control back to the user.

6. From Design to Implementation

Translating UX into product rules

Matched

Eligible

Prepare

Approval / Auto Submit

Submitted

Failed → Recovery

Together with engineering, we translated these interaction flows into implementable state logic covering asynchronous execution, third-party constraints, approval rules, and recovery scenarios.

7. Outcome

Autonomy-specific validation signals

Auto Apply adoptionApproval rateSuccessful submission rateManual intervention rate

8. Key Takeaway
The most valuable design decision was not increasing automation, but allowing users to define when automation was appropriate.

Design Decision 04

Application Accountability

Making AI actions transparent

How do users stay informed after AI takes action on their behalf?

1. The Design Challenge

Once applications could be submitted automatically, the interaction no longer ended when the user clicked Apply. Users needed to understand what had happened, what was still pending, and whether anything required their attention.

Automation without visibility quickly becomes a black box.

2. What We Learned

The anxiety was not simply whether an application existed. Users needed to verify the outcome, the materials submitted, any status change, and whether the next step belonged to them.

Did it go through?Which resume was submitted?Has anything changed?Do I need to act?

Users wanted confidence after automation, not just confirmation that it had started.

3. Design Decision

We made every automated action traceable.

Rather than treating automation as a background process, we designed every AI action to remain visible, reviewable, and actionable after execution.

Before

After

Submission disappears after execution

Every application remains traceable

Only success notifications

Complete application history

AI acts in the background

AI actions stay visible

Early direction

A dashboard listed submitted roles and their latest status.

Gap

Status alone did not reveal which material was sent, why an application failed, or where a recruiter reply belonged.

Final

Connected the dashboard and unified inbox to one traceable application detail view.

4.1

Every application has a visible history

The complete journey—from discovery and submission to recruiter response—remains reviewable.

4.2

Status communicates progress, not just outcomes

Users can see what is complete, what is still in progress, and what changed along the way.

4.3

Surface actions that require attention

The system interrupts users only when approval, recovery, or a timely response is needed.

5. Translating the Decision into the Experience

Visibility continues after the AI acts.

A. Application Timeline

Progress stays legible

Users could always understand where each application stood—not just its latest outcome.

Applied

Materials submitted

Under Review

Recruiter reviewing

Interview

Response required

Offer

Decision pending

Rejected

Journey recorded

JobNova application detail showing a chronological application timeline, submitted materials, status changes, and recruiter email. — B. Submission Record
Submitted resume, cover letter, date, match evidence, and the Auto Apply log remained reviewable after completion.

JobNova inbox classifying application updates and highlighting messages that require the user to respond. — C. Actionable Notifications
Interview invitations, failures, approvals, and material updates surfaced only when user attention was required.

6. Outcome

Accountability-specific validation signals

Application tracking usageNotification open rateTime to respondInterview follow-up completion

7. Key Takeaway
Automation becomes trustworthy only when users can understand what happened, what is happening, and what needs their attention next.

Human–AI Collaboration Model

Four decisions, one trust model

The four experiences form one continuous relationship: AI explains, collaborates, acts within permission, and remains accountable afterward.

Recommendations

Explain

AI makes its reasoning understandable.

Authorship

Collaborate

AI proposes; users approve.

Autonomy

Act

AI works within user-defined boundaries.

Accountability

Account

Every AI action stays visible and traceable.

8. Design to Launch

The handoff needed product rules, not just screens.

JobNova depended on matching logic, AI generation, auto-submission states, third-party platform constraints, and feedback loops. I worked with engineering to turn the experience into implementable system behavior.

Handoff Scope

Feature priority

Page and component states

AI generation flow

Application submission states

Error and recovery logic

Third-party platform constraints

Tracking and launch QA

Design System Coverage

Typography and color tokensButtons and form componentsJob and application cardsMatch and application statusAI content highlightsSuccess, warning, and error feedbackNotification and confirmation patterns

Design system principle
I treated the design system as a set of intentional constraints, not a catalogue to expand. For every new interaction, I first reused an existing token or component. If it could not meet the need, I created a controlled variant; only when neither option preserved the intended behavior did I introduce something new.

9. Results

After launch, we validated the product through behavior, reliability, and outcome quality.

The core question was not only how many applications users submitted. We looked at whether users adopted controlled automation, whether the system executed reliably, and whether high-match applications produced meaningful responses.

Activation

68.06%

30-day onboarding completion

179 of 263 new users uploaded a resume and set job preferences. This is setup completion, not “finished a first job match.”

Adoption

19.39%

30-day Auto Apply activation

51 of 263 new users enabled Auto Apply. Among the 179 who finished onboarding, that is 28.5%.

Reliability

88.21%

Final application success rate

1,803 automated applications submitted in 30 days, after queues and retries. It measures execution, not user trust.

Speed

1.03 hrs

Match-to-application time

Average time from match generation to successful auto-submit, including queues and retries.

Outcome

10.77%

Interview invitation rate

7 of 65 successful Auto Apply users. Inbox-classified, n=65. Directional only.

10. Reflection

Closing perspective

Designing AI products was never about increasing automation.
It was about making automation understandable, configurable, and accountable.

Three learnings I will carry forward

Understand → Configure → Account

Understandable

Trust begins before the AI acts

A recommendation or generated result becomes useful only when users can understand the evidence, uncertainty, and changes behind it.

Configurable

Control is a system, not a switch

Users need to define the conditions, permissions, and review boundaries under which AI is allowed to represent them.

Accountable

Automation continues after execution

Every action must remain visible, traceable, and recoverable so users always know what happened and what requires attention next.

JobNova AI Job Search Platform

AI tools optimized job-search tasks.The journey stayed fragmented.

The work ran as a loop, not a handoff.Discover, reframe with the team, then review what shipped.

Early Opportunity

Discovery & Scoping

Define & Design

Build

QA & Review

Launch

Post-Launch

We assumed AI could replace pieces of the job-search journey.

AI made parts of applying faster.

The journey still ran on manual work.

Matching and writing were not the bottleneck.Users needed applying to happen without them.

Users would focus on interviews.The product had to apply.

The challenge was not whether to automate applying.It was how to balance autonomy and control.

Applying on someone’s behalf is the highest-stakes moment in the loop.

User Experience Designer

User Experience Designer

If an application disappears after send, the automation has failed.

History

Status

Attention

The four moments had to feel like one product.

The goal was never maximum automation.It was automation people could understand, configure, and hold accountable.

30-day onboarding completion

30-day Auto Apply activation

Final application success rate

Match-to-application time

Interview invitation rate

What experience can still raise

30-day onboarding completion

30-day Auto Apply activation

Three learnings I will carry forward

Explain before acting

Make control configurable

Design beyond execution

User Experience Designer

User Experience Designer

AI tools optimized job-search tasks.
The journey stayed fragmented.

The work ran as a loop, not a handoff.
Discover, reframe with the team, then review what shipped.

Matching and writing were not the bottleneck.
Users needed applying to happen without them.

Users would focus on interviews.
The product had to apply.

The challenge was not whether to automate applying.
It was how to balance autonomy and control.

The goal was never maximum automation.
It was automation people could understand, configure, and hold accountable.