Daphne Cheng

Daphne Cheng

Daphne Cheng

Contents

Overview

Problem

Discovery

Design

Conclusion

Contents

Overview

Problem

Discovery

Design

Conclusion

Explore Careers: AI Chatbot to Find Career Paths

Designing an AI chatbot for high school students to explore and find career paths

Project Type

Conversational AI, UX Design, User Research

My Roles

UX Content Designer, Conversation Architect, Prototyper

Team

Design Lead (me), 2 Researchers, 1 AI Engineer, 4 Designers

Duration

9 Months

Situation

Too many choices, too little time

There are 2000+ careers. How is a 16-year-old supposed to choose? Most students don’t have access to in-depth counseling, and even if they do, it's brief and surface-level. The earlier students get direction, the better— but not everyone has the same resources.

There are 2000+ careers. How is a 16-year-old supposed to choose? Most students don’t have access to in-depth counseling, and even if they do, it's brief and surface-level. The earlier students get direction, the better— but not everyone has the same resources.

There are 2000+ careers. How is a 16-year-old supposed to choose? Most students don’t have access to in-depth counseling, and even if they do, it's brief and surface-level. The earlier students get direction, the better— but not everyone has the same resources.

Task

Career help that doesn't suck

Students deserve better than random career quizzes. They need real advice, real insights, real guidance. A 24/7 career mentor that actually spend the time to understand their strengths. A accessible personalized career help.

Students deserve better than random career quizzes. They need real advice, real insights, real guidance. A 24/7 career mentor that actually spend the time to understand their strengths. A accessible personalized career help.

Students deserve better than random career quizzes. They need real advice, real insights, real guidance. A 24/7 career mentor that actually spend the time to understand their strengths. A accessible personalized career help.

Action

Counselor that is made for you

We built an AI chatbot using:

  • GPT 3.0 LLM fine-tuned for career advice

  • NLP and machine learning to personalize responses

  • Live job data APIs (BLS.gov, O*NET, SerpAPI)

We layered in:

  • Live goal setting sessions with ai writing into the Google Doc with you

  • Friendly, motivational tone and smart conversation flows

  • Realistic wage and education data with citations

We built an AI chatbot using:

  • GPT 3.0 LLM fine-tuned for career advice

  • NLP and machine learning to personalize responses

  • Live job data APIs (BLS.gov, O*NET, SerpAPI)

We layered in:

  • Live goal setting sessions with ai writing into the Google Doc with you

  • Friendly, motivational tone and smart conversation flows

  • Realistic wage and education data with citations

We built an AI chatbot using:

  • GPT 3.0 LLM fine-tuned for career advice

  • NLP and machine learning to personalize responses

  • Live job data APIs (BLS.gov, O*NET, SerpAPI)

We layered in:

  • Live goal setting sessions with ai writing into the Google Doc with you

  • Friendly, motivational tone and smart conversation flows

  • Realistic wage and education data with citations

Results

Impact metrics

My team created a chatbot model that will seek to understand the student first, dig for students' strengths, and provide professional advice based on actual web results. We make sure no one feels lost. Turning confusion into confidence.

My team created a chatbot model that will seek to understand the student first, dig for students' strengths, and provide professional advice based on actual web results. We make sure no one feels lost. Turning confusion into confidence.

My team created a chatbot model that will seek to understand the student first, dig for students' strengths, and provide professional advice based on actual web results. We make sure no one feels lost. Turning confusion into confidence.

Challenge

Build a reliable chatbot for career counseling that doesn't hallucinate

Results

🚀 85% of students said the chatbot helped more than a human counselor

🎯 3+ students per school walked away career-ready.

5 min read

Case Study

continue scrolling

The problem

Confusion

High school students often struggle with choosing a career path due to limited guidance, overwhelming information, and a lack of personalized resources.

High school students often struggle with choosing a career path due to limited guidance, overwhelming information, and a lack of personalized resources.
High school students often struggle with choosing a career path due to limited guidance, overwhelming information, and a lack of personalized resources.

How can we transform the inaccessible & generic traditional career counseling experience into LESS CONFUSING and MORE DECISIVE recommendations?

How Might We
make it LESS OVERWHELMING to navigate academics, culture, and daily life for exchange students?

…with a highly personalized AI!

Not to be blunt, but human cost are high— most school districts simply don't have the funding to provide 1:1 career guidance for students, if any at all. Which is why I made it my team's goal to build a conversational AI chatbot that provides accessible tailored career guidance by analyzing students’ interests, strengths, and goals, while preserving the human side of exploring and figuring out one's future. We want to set students up and take away the human cost, so students can find the resources they need easier. It offers:

  • Conversational assessments to identify strengths and major/career matches.

  • College and vocational path suggestions based on information told by student, job market trends from the https://www.bls.gov/, and accurate numbers for wage and predicted years of education and cost

  • Goal setting, ask student to take action, and set possible career and college path that is personalized to students needs in short and long term form

Finding our target audience

We conducted surveys on a focus groups with local high school to understand their expectations from a career advisor.

High school students (ages 14-18)

  • Understands technology but not always aware of professional tools

  • Need engaging, bite-sized career advice

  • Often anxious about the future and need reassurance

  • want actionable organizaitons/actions they can do right now

1/ Overwhelming career choices

2/ Generic school career guidance

3/ Uncertainty about community college vs university, and which majors to pick

4/ Lack of personalized insights based on what they want to do, what they think is fun, and what's realistic for them

Discovery

Tone & voice

Based on our findings, students are more willing to share and provide better understanding of what skills they have and what position they stand through conversational flow. Using on a casual tone takes the effort away from needing to structure their thoughts into well worded sentences. Friendly, relatable, and slightly informal to make the chatbot feel like a supportive mentor. Encouraging and motivational to ease anxiety about future decisions.

Based on our findings, students are more willing to share and provide better understanding of what skills they have and what position they stand through conversational flow. Using on a casual tone takes the effort away from needing to structure their thoughts into well worded sentences. Friendly, relatable, and slightly informal to make the chatbot feel like a supportive mentor. Encouraging and motivational to ease anxiety about future decisions.

Based on our findings, students are more willing to share and provide better understanding of what skills they have and what position they stand through conversational flow. Using on a casual tone takes the effort away from needing to structure their thoughts into well worded sentences. Friendly, relatable, and slightly informal to make the chatbot feel like a supportive mentor. Encouraging and motivational to ease anxiety about future decisions.

Here is our basic conversation flow

Introduction & Icebreaker – Engaging first message to establish trust.

  1. Interest Assessment – Fun quizzes and quick personality tests to determine career preferences.

  2. Personalized Career Suggestions – Providing multiple paths with pros and cons.

  3. Interactive Guidance – Students can ask follow-up questions about salary expectations, job outlook, necessary education, etc.

  4. Next Steps – Offering actionable advice like internship opportunities, online courses, and college and major choices.

UI needs for the chatbot interface

Simple, chatbot-first interface for quick interactions

  • integration to current explore careers platform

  • aesthetic consistency

  • big enough to fit long llm response

Design & Development

Tech Stack

  • Database: https://www.bls.gov/, o*net https://www.onetonline.org/

  • AI Model: 3.o GPT-based fine-tuned LLM for career advice generation

  • API: SerpAPI to integrate web search

  • Prototyping and testing: Figma to collect analytics and make quick iterations for research

  • Database: https://www.bls.gov/, o*net https://www.onetonline.org/

  • AI Model: 3.o GPT-based fine-tuned LLM for career advice generation

  • API: SerpAPI to integrate web search

  • Prototyping and testing: Figma to collect analytics and make quick iterations for research

  • Database: https://www.bls.gov/, o*net https://www.onetonline.org/

  • AI Model: 3.o GPT-based fine-tuned LLM for career advice generation

  • API: SerpAPI to integrate web search

  • Prototyping and testing: Figma to collect analytics and make quick iterations for research

User Interviews to see what users see

blahblah our very long video calls on zoom with like 20 high schoolers, each one at least 2 times

snippet from our interview doc

1/ Generic career advice

⚠️ Issue
It kept giving the same advice to users with drastic different career tracks (go to college, get a degree, find an internship— even when we ask if we want to be a clown) (recommended vocational school but then also asked me to find an internship— which I guess clown people don't actually have cuz we can;t find any info regarding that)

❓Why
The version we are using, GPT 3.o, repeats a lot of past conversations, which means the advice comes out to be generic

🛠️ Fix
Adjusted metrics on the bot setting to remember less about the conversation, and increased the creativity slider so answers can have a wider range

before and after layout redesign

2/ Keeping students engaged

⚠️ Issue
Students lose interest quickly in the goal setting function of our chatbout, resulting in less updates on their personal goals (not even going more than 3 days) and not giving their latest information, so chatbot was giving advice based on old understanding of user

❓Why
In the end, this IS an AI bot. Students don't really want to spend an afternoon just talking with a bot. there's nothing physical to keep them accountable

🛠️ Fix
Re-centered goal setting prompts to be more instrintic motivation— more goals about self, physically asking the users to track it themselves. also a google docs thingy. we got gpt to write users' goals into a google doc, so they can go in and edit it physically.

booking icon is moved from top bar to bottom nav bar

3/ hallucinations

⚠️ Issue
the bot was spewing random numbers for wage, and lots of rEnsuring reliability of career data and compliance with educational standards

❓Why
model likes to hallucinate about numbers, and match unrelated results with careers, as well as say the same stuff from previous career to new careers

🛠️ Fix
changed prompt so can only take numbers quotes from database, but not generating a number iself. and must include a reference link for where that number came from

event card has "star" function now

Conclusion

Making things more accessible

with the rise of ai back in 2023, my lab really saw this as an opportunity to bridge the gap in education resources. AI let us lower the human cost of career counseling and democratize access to high-quality guidance. We replaced a luxury service with something scalable, affordable, and just as helpful.

with the rise of ai back in 2023, my lab really saw this as an opportunity to bridge the gap in education resources. AI let us lower the human cost of career counseling and democratize access to high-quality guidance. We replaced a luxury service with something scalable, affordable, and just as helpful.

with the rise of ai back in 2023, my lab really saw this as an opportunity to bridge the gap in education resources. AI let us lower the human cost of career counseling and democratize access to high-quality guidance. We replaced a luxury service with something scalable, affordable, and just as helpful.

Results

out of the 15 group of active participants thought a 3 month tracking period, 3 students stayed with us till the end and was able to apply to the colleges and majors they want and like and are hearing back for their results. i believe they walked away with goals in their mind,

Reflections

throughout this 9 month journey, what surprised me the most is how motivated students actually are. i think our little research showed that with even just a little motivation and a more conversaitonal way to show the how to. by presenting out of reach information through a conversaitonal tone, teenagers have the abilities to find what they want and from there on do what they need to do.

i think its a good call that we pivoted from replicating a human counselor to focus on goal setting and giving accurate information in a conversaiotnal way.

What's next?

  • Labeling final prototypes for dev handoff

  • Exploring GPT-4 for better consistency

  • Localizing data storage for longer goal tracking (beyond 3 months