Career Highways AI Integration
Career Highways AI Integration
Redesigning Career Exploration Experience by Integrating Artificial Intelligence
Redesigning Career Exploration Experience by Integrating Artificial Intelligence
CONTENT
OVERVIEW
OVERVIEW
Career Highways is a web platform that helps high school students explore and plan career pathways. During my internship at PUSH Studio, I explored how AI could enhance the Explore experience while preserving familiar search behaviors.
As a UX/UI Designer, I prototyped two AI interaction models, a conversational chatbot and a recommendation based system, and conducted usability testing to evaluate when AI supports exploration versus traditional search.
Career Highways is a web platform that helps high school students explore and plan career pathways. During my internship at PUSH Studio, I explored how AI could enhance the Explore experience while preserving familiar search behaviors.
As a UX/UI Designer, I prototyped two AI interaction models, a conversational chatbot and a recommendation based system, and conducted usability testing to evaluate when AI supports exploration versus traditional search.
Team:
Team
Team
I collaborated with an AI Engineer, PUSH Design Director, and the Client.
I collaborated with an AI Engineer, PUSH Design Director, and the Client.
Methods:
Methods
Methods
Iterative prototyping, and usability testing.
Iterative prototyping, and usability testing.
Tools:
Tools
Tools
Figma, FigJam, Jira, and Zoom.
Figma, FigJam, Jira, and Zoom.
Duration:
Duration
Duration
5 weeks
5 weeks
Impact
Impact
The project revealed that exploration and precision search are distinct behaviors that benefit from different interaction models. These insights showed how AI assisted exploration and traditional search can coexist, and revealed an opportunity to consolidate four Explore pages into a single discovery experience.
The project revealed that exploration and precision search are distinct behaviors that benefit from different interaction models. These insights showed how AI assisted exploration and traditional search can coexist, and revealed an opportunity to consolidate four Explore pages into a single discovery experience.
PROBLEM SPACE
PROBLEM SPACE
Introducing AI without disrupting familiar search behavior became the core design challenge.
Introducing AI without disrupting familiar search behavior became the core design challenge.
The Career Highways Explore experience already supported discovery through keyword search and traditional filters. The challenge was integrating AI in a way that felt intuitive while minimizing disruption to an established workflow.
The Career Highways Explore experience already supported discovery through keyword search and traditional filters. The challenge was integrating AI in a way that felt intuitive while minimizing disruption to an established workflow.
Central Question:
How might we integrate AI into the Explore experience in a way that feels intuitive to users while minimizing disruption to the existing workflow?
Central Question:
How might we integrate AI into the Explore experience in a way that feels intuitive to users while minimizing disruption to the existing workflow?
CONTEXT
CONTEXT
The existing Explore experience fragmented discovery across multiple pages.
The existing Explore experience fragmented discovery across multiple pages.
The Explore experience helps students discover career paths, jobs, certifications, and courses. Before this project, these categories were separated across four different Explore pages. Users often repeated similar searches when moving between them.
The Explore experience helps students discover career paths, jobs, certifications, and courses. Before this project, these categories were separated across four different Explore pages. Users often repeated similar searches when moving between them.
The project began with a request to introduce prominent AI capabilities while maintaining familiarity for returning users, which introduced several design constraints throughout the project including:
The project began with a request to introduce prominent AI capabilities while maintaining familiarity for returning users, which introduced several design constraints throughout the project including:
AI visibility versus screen space
AI visibility versus screen space
Prominent chatbot layouts reduced space for browsing results.
Prominent chatbot layouts reduced space for browsing results.
Fragmented exploration
Fragmented exploration
Related content lived across multiple pages, increasing mental effort.
Related content lived across multiple pages, increasing mental effort.
Overlapping systems
Overlapping systems
AI, keyword search, and filters appearing together made system behavior unclear.
AI, keyword search, and filters appearing together made system behavior unclear.
EARLY DIRECTION
EARLY DIRECTION
Early concepts explored conversational AI as the primary exploration tool.
Early concepts explored conversational AI as the primary exploration tool.
Based on leadership direction, I designed an AI-first happy path with a conversational interface as the focal point. I explored direction through rapid high-fidelity iteration, using the flow to surface early tensions between AI prominence and result visibility rather than treating it as a final design.
The initial design included the following:
Based on leadership direction, I designed an AI-first happy path with a conversational interface as the focal point. I explored direction through rapid high-fidelity iteration, using the flow to surface early tensions between AI prominence and result visibility rather than treating it as a final design.
The initial design included the following:
Early feedback focused on reducing redundancy across search controls while reinforcing AI visibility.
Early feedback focused on reducing redundancy across search controls while reinforcing AI visibility.
AI VS. CLASSIC SEARCH
AI VS. CLASSIC SEARCH
Separating AI and classic search clarified how users entered the experience.
Separating AI and classic search clarified how users entered the experience.
As the AI Engineer joined the project, concerns surfaced about users being unintentionally pushed into AI. Through multiple reviews, it became clear that combining AI conversation, keyword search, and filters in one flow created confusion about system behavior.
In response, I redesigned the interface to clearly separate AI-assisted exploration from classic keyword search, allowing users to intentionally choose how they wanted to interact.
As the AI Engineer joined the project, concerns surfaced about users being unintentionally pushed into AI. Through multiple reviews, it became clear that combining AI conversation, keyword search, and filters in one flow created confusion about system behavior.
In response, I redesigned the interface to clearly separate AI-assisted exploration from classic keyword search, allowing users to intentionally choose how they wanted to interact.
Based on feedback, I removed filter chips and kept collapsible filters behind both modes to reduce visual noise.
Based on feedback, I removed filter chips and kept collapsible filters behind both modes to reduce visual noise.
MODE SWITCH
MODE SWITCH
Introducing a toggle allowed users to switch exploration modes without losing context.
Introducing a toggle allowed users to switch exploration modes without losing context.
Further feedback showed users needed flexibility to switch modes mid-exploration. I introduced a persistent toggle that allowed users to move between AI and classic search without losing context.
Further feedback showed users needed flexibility to switch modes mid-exploration. I introduced a persistent toggle that allowed users to move between AI and classic search without losing context.
RECOMMENDATION MODEL
RECOMMENDATION MODEL
A recommendation model revealed an opportunity to consolidate four separate Explore pages into one experience.
A recommendation model revealed an opportunity to consolidate four separate Explore pages into one experience.
The recommendation model began as a design judgment call, rather than a fully validated direction. I questioned whether a chatbot alone supported early-stage exploration, especially for high school students who may not yet know what to search for.
Instead of treating this as a final solution, I introduced the recommendation model as a testable alternative. Its purpose was to explore whether a more passive, content-driven experience felt easier or more engaging than conversational input
The recommendation model design included the following:
The recommendation model began as a design judgment call, rather than a fully validated direction. I questioned whether a chatbot alone supported early-stage exploration, especially for high school students who may not yet know what to search for.
Instead of treating this as a final solution, I introduced the recommendation model as a testable alternative. Its purpose was to explore whether a more passive, content-driven experience felt easier or more engaging than conversational input
The recommendation model design included the following:
This concept also revealed an opportunity to surface related careers, jobs, certifications, and courses within a single discovery experience.
This concept also revealed an opportunity to surface related careers, jobs, certifications, and courses within a single discovery experience.
USER TESTING
USER TESTING
User testing revealed different AI models support different exploration behaviors.
User testing revealed different AI models support different exploration behaviors.
I conducted moderated usability testing with 5 participants, each testing both AI models. The goal of testing was to understand user preference between conversational AI and recommendation-based exploration, and to evaluate how each model affected clarity, engagement, and perceived usefulness.
I conducted moderated usability testing with 5 participants, each testing both AI models. The goal of testing was to understand user preference between conversational AI and recommendation-based exploration, and to evaluate how each model affected clarity, engagement, and perceived usefulness.
Findings
Findings
Users preferred AI for exploration and classic search for precise goals.
Users preferred AI for exploration and classic search for precise goals.
Overlapping AI and classic search caused confusion.
Overlapping AI and classic search caused confusion.
Fixed chatbot placement disrupted browsing.
Fixed chatbot placement disrupted browsing.
Users often missed when AI results updated.
Users often missed when AI results updated.
Key Takeaway:
Exploration and precision search are distinct behaviors that benefit from different AI models.
Key Takeaway:
Exploration and precision search are distinct behaviors that benefit from different AI models.
Limitation
Limitation
Non-representative Sample
Non-representative Sample
Participants were computing students comfortable with AI, which may have influenced adoption. Testing with high school students is needed to validate findings.
Participants were computing students comfortable with AI, which may have influenced adoption. Testing with high school students is needed to validate findings.
FINAL DESIGN
FINAL DESIGN
Combining AI exploration and traditional search created a more flexible discovery experience.
Combining AI exploration and traditional search created a more flexible discovery experience.
Testing showed that no single AI model supported every exploration behavior. Instead of choosing one direction I recommended a combined approach.
Testing showed that no single AI model supported every exploration behavior. Instead of choosing one direction I recommended a combined approach.
Key Design Changes
Key Design Changes
Simplified Layout
Simplified Layout
Removed the hero banner and consolidated Explore pages.
Removed the hero banner and consolidated Explore pages.
Preserved keyword search and filters in classic mode.
AI mode removes traditional filters and surfaces recommendations.
AI mode removes traditional filters and surfaces recommendations.
Guided Exploration
Guided Exploration
Sticky conversation window for reference.
Sticky conversation window for reference.
Visual cues and summaries signal result updates.
Visual cues and summaries signal result updates.
Recommendation carousels support browsing and discovery.
Recommendation carousels support browsing and discovery.
This approach consolidated fragmented exploration into a single experience and separated AI from classic search, clarifying how the system works while preserving familiar search behavior.
This approach consolidated fragmented exploration into a single experience and separated AI from classic search, clarifying how the system works while preserving familiar search behavior.
CONCLUSION
CONCLUSION
Separating AI exploration and traditional search created a clearer discovery experience.
Separating AI exploration and traditional search created a clearer discovery experience.
Impact
Impact
This project showed how AI can be integrated into an existing discovery experience without disrupting familiar search behaviors. Testing revealed that exploration and precision search are distinct behaviors, which informed separating AI assisted exploration from traditional search. Exploring a recommendation based model also revealed an opportunity to consolidate four Explore pages into a single discovery experience.
This project showed how AI can be integrated into an existing discovery experience without disrupting familiar search behaviors. Testing revealed that exploration and precision search are distinct behaviors, which informed separating AI assisted exploration from traditional search. Exploring a recommendation based model also revealed an opportunity to consolidate four Explore pages into a single discovery experience.
Retrospective
Retrospective
If I were to approach this project again, I would invest more time in early user research and competitive analysis to clarify user goals before designing AI solutions. Although I raised layout concerns early and suggested a collapsible AI pattern, I wish I had more time and stronger influence to communicate the tradeoffs and validate alternative layouts through testing. This may have led to a solution that better balanced AI visibility and browsing efficiency.
If I were to approach this project again, I would invest more time in early user research and competitive analysis to clarify user goals before designing AI solutions. Although I raised layout concerns early and suggested a collapsible AI pattern, I wish I had more time and stronger influence to communicate the tradeoffs and validate alternative layouts through testing. This may have led to a solution that better balanced AI visibility and browsing efficiency.