CareerQuest

AI-powered career exploration and skill-building platform for high school and college students.

CareerQuest helps young learners navigate the overwhelming landscape of career options through a multi-agent AI pipeline, gamified skill trees, and personalized learning roadmaps — transforming career discovery from a daunting task into an engaging, guided journey.

Hackathon: Hack The Experience (HTE) 2026 Team: DreamHackers


Problem Statement

1. Information Overwhelm

High school students face an avalanche of career information — job boards, course platforms, aptitude tests, college guides — without the experience or mentorship to filter what matters. Most give up exploring or pick paths based on peer pressure rather than genuine fit.

2. AI Is Reshaping the Job Market

Artificial intelligence is automating roles faster than the education system can adapt. Students choosing careers today need to understand which fields are AI-resilient and which face disruption — information that traditional guidance counselors rarely provide.

3. Engagement Gap

Existing career tools are static, text-heavy, and forgettable. Students need the same engagement hooks that keep them on Duolingo and TikTok — XP progression, visual skill trees, achievement badges, and bite-sized interactions — applied to career planning.


Solution

CareerQuest combines a multi-agent AI pipeline with a gamified learning interface to deliver:

  • Personalized career matching — NLP parses free-text input, vector search finds semantically similar careers from 1,000+ occupations, and an LLM evaluator selects the top 3 based on fit, growth outlook, AI resilience, and salary trajectory.
  • 4-horizon learning roadmaps — From "Now → End of High School" through "Post-Grad → 5 Years", each career gets a time-phased action plan with matched courses.
  • AI automation risk scoring — Every career has a 0–100 risk score blending O*NET structural data with 5 live market signals (academic research, news, job postings, ETF trends, social media sentiment).
  • Gamified skill tree — 8 career realms, 300+ unlockable skill nodes, XP progression, achievement badges, and visual unlock effects.
  • Duolingo-style UX — Clean, responsive UI with dark/light mode, scroll-based engagement, particle effects on skill completion, and a guided onboarding questionnaire.

Architecture

┌─────────────────────────────────────────────────────────┐
│                      Frontend                           │
│   React 18 · TypeScript · Tailwind CSS · shadcn/ui      │
│   Vite · React Router v6 · TanStack Query               │
└─────────────────┬───────────────────────────────────────┘
                  │ REST API
┌─────────────────▼───────────────────────────────────────┐
│                    Backend API                          │
│         FastAPI · Pydantic v2 · Uvicorn                 │
└─────────────────┬───────────────────────────────────────┘
                  │
┌─────────────────▼───────────────────────────────────────┐
│           Multi-Agent Pipeline (LangGraph)               │
│                                                         │
│   ┌─────┐   ┌──────────┐   ┌───────────┐   ┌────────┐  │
│   │ NLP │──▶│  Vector   │──▶│  Career   │──▶│Learning│  │
│   │Agent│   │  Search   │   │ Evaluator │   │  Path  │  │
│   └─────┘   └──────────┘   └───────────┘   └────────┘  │
│                                                         │
│   Fault-Tolerant LLM Chain:                             │
│   Primary:   Minimax API (MiniMax-Text-01)              │
│   Secondary: AWS Bedrock (Minimax)                      │
│   Fallback:  OpenRouter (gpt-oss-120b)                  │
└─────────────────┬───────────────────────────────────────┘
                  │
┌─────────────────▼───────────────────────────────────────┐
│               Data Layer                                │
│   Supabase (PostgreSQL + pgvector + Auth)                │
│   1,000+ career embeddings (text-embedding-3-small)     │
│   O*NET · BLS · Frey-Osborne · Live Market Signals      │
└─────────────────────────────────────────────────────────┘

Multi-Agent Pipeline

The career advisor uses a 5-node LangGraph state machine that processes user input through successive stages of refinement:

Stage Agent Input Output
1 NLP Agent Free-text user query ParsedUserInput — goals, skills, interests, hobbies, constraints
2 Vector Search Embedded user profile Top 10 semantically matched careers from pgvector
3 Career Evaluator 10 candidates + student profile Top 3 careers with scores, rationale, skills to develop
4 Bridge Evaluator output Normalized JobRecommendation objects
5 Learning Path Selected careers + student context 4-horizon personalized roadmap with matched courses

AI Techniques

Technique Implementation
NLP LLM-based structured extraction — parses natural text into typed Pydantic models (goals, skills, interests, constraints)
RAG Career descriptions embedded via text-embedding-3-small into pgvector; cosine similarity retrieval augments LLM prompts with real career data
LangChain/LangGraph Orchestrates the multi-agent pipeline as a typed state graph with instrumented nodes and edge transitions
Multimodal Forecasting AI risk scores blend structural O*NET data with 5 live signals — academic papers (Semantic Scholar), news sentiment (GDELT), job postings (Indeed), sector ETF trends (yfinance), social media sentiment (Reddit)
Rubric-LLM Prompting Career evaluator uses a 5-dimension rubric (fit, growth, AI resilience, salary, accessibility) for structured scoring rather than free-form generation

Fault-Tolerant LLM Chain

Request ──▶ Minimax API (MiniMax-Text-01)
               │
               ▼ (on failure)
            AWS Bedrock (Minimax)
               │
               ▼ (on failure)
            OpenRouter (gpt-oss-120b)

The pipeline implements cascading LLM failover to ensure uptime. If the primary Minimax API is unavailable, requests fall back to AWS Bedrock (which hosts Minimax models), and finally to OpenRouter as a last resort. All three providers use the same structured output interface, making failover transparent to the pipeline.


Tech Stack

Layer Technology
Frontend React 18, TypeScript, Vite, Tailwind CSS, shadcn/ui, React Router v6, TanStack Query, Recharts
Backend API FastAPI, Pydantic v2, Uvicorn, Python 3.13
AI Pipeline LangGraph, LangChain, Minimax API, AWS Bedrock, OpenRouter
Embeddings text-embedding-3-small via OpenRouter (1024 dimensions)
Vector DB Supabase pgvector (cosine similarity search)
Database Supabase (PostgreSQL + Row-Level Security + Auth)
Data Sources O*NET (1,000+ occupations), BLS wages, Frey-Osborne automation scores, Semantic Scholar, GDELT, yfinance, Reddit
Infrastructure Terraform (AWS EC2 + ECR), Docker, GitHub Actions CI/CD
AWS Services Bedrock (LLM failover), Bedrock AgentCore (evaluation + memory), EC2, ECR

Key Features

AI Career Advisor

Chat interface that analyzes student interests, skills, and goals through natural language. Returns 3 personalized career recommendations with match scores, skill gap analysis, AI risk outlook, and growth projections.

Gamified Skill Tree

Interactive node graph spanning 8 career realms. Students unlock skills in a hierarchical tree, earning XP and progressing through levels (Explorer → Apprentice → Specialist → Expert → Master → Veteran → Champion → Legend). Features pan & zoom navigation, particle effects on unlock, sound effects, and visual completion badges.

Personalized Roadmaps

Each recommended career gets a 4-horizon learning roadmap:

  • Now → End of High School — foundational skills, AP courses, extracurriculars
  • College (Year 1–2) — core courses, introductory tools, clubs & research
  • College (Year 3–4) — specialization, internship targets, capstone projects
  • Post-Graduation → 5 Years — entry roles, certifications, skill deepening

Each roadmap item is enriched with up to 3 matched courses from the platform database.

AI Automation Risk Scoring

Every career has a composite risk score (0–100) computed as:

  • 50% Structural baseline — O*NET Work Activities + Abilities scored against Frey-Osborne automation classifications
  • 50% Live signals — 5 real-time indicators weighted and evaluated per domain:
    • Academic research volume (30%) — Semantic Scholar API
    • Job posting trends (25%) — Indeed Hiring Lab
    • News sentiment (20%) — GDELT
    • Sector ETF divergence (15%) — yfinance
    • Social media sentiment (10%) — Reddit API

Achievements & Analytics

XP-based leveling system with milestone badges (First Steps, Skill Hunter, Pathfinder, Domain Completionist), realm progress visualization, and learning analytics dashboard.

Onboarding Flow

5-step interactive questionnaire with mascot guidance:

  1. Activities — 12 activity categories (coding, art, science, music, etc.)
  2. Interests — Multi-select from 20 interest areas (AI, Robotics, Economics, etc.)
  3. Learning Style — Quiz on preferred learning method, pace, and goals
  4. Confidence — Self-assessment slider
  5. Grade Level — 9th grade through college sophomore

Skill Realms

ID Realm Description
1 Technology Software development, data science, cybersecurity
2 Healthcare Medicine, nursing, health sciences
3 Finance & Business Accounting, management, marketing
4 Engineering Mechanical, civil, electrical, chemical
5 Arts & Design Graphic design, UX/UI, media production
6 Trades & Construction Skilled trades, manufacturing, logistics
7 Education Teaching, counseling, training
8 Science & Research Biology, chemistry, physics, environmental science

Quick Start

Prerequisites

  • Node.js v18+ and npm
  • Python 3.13+ and uv
  • Docker (optional, for containerized backend)
  • Supabase project with pgvector enabled
  • API keys: Minimax, OpenRouter, Supabase

Backend

cd backend
cp .env.example .env   # Edit with your API keys
uv sync
uv run uvicorn app.main:app --host 0.0.0.0 --port 8000

Frontend

cd frontend
cp .env.example .env   # Edit VITE_BACKEND_URL
npm install
npm run dev

Frontend: http://localhost:5173 | Backend API: http://localhost:8000

Docker (Backend)

cd backend
docker compose up --build

Data Pipeline

Seed the career database with 1,000+ occupations:

cd backend

# 1. Download O*NET + BLS raw data
uv run python -m agents.scripts.download_onet_bls

# 2. Transform into careers_onet.json
uv run python -m agents.scripts.build_careers_from_onet

# 3. Embed and upsert to pgvector
uv run python -m agents.scripts.seed_vector_db

# 4. (Optional) Refresh AI risk scores with live signals
uv run python -m agents.scripts.update_ai_risk

Environment Variables

Backend

Variable Description
SUPABASE_URL Supabase project URL
SUPABASE_ANON_KEY Supabase anon/public key
SUPABASE_DB_URL PostgreSQL connection string
OPENROUTER_API_KEY OpenRouter API key (LLM + embeddings)
MINIMAX_API_KEY Minimax API key (primary LLM)
CORS_ORIGINS Allowed CORS origins (default: *)

Frontend

Variable Description
VITE_BACKEND_URL Backend API URL (e.g., http://localhost:8000)
VITE_WEBSITE_API_URL Course platform API URL
VITE_WEBSITE_URL Course platform URL

API Endpoints

Method Path Auth Description
POST /auth/signup No Register a new user
POST /auth/signin No Login, returns JWT
POST /auth/logout Yes Invalidate session
POST /career-advisor Yes Full career advisor pipeline
POST /career-advisor/cards Yes Career cards only (no roadmaps)
POST /career-advisor/roadmap Yes Generate roadmaps for selected careers
GET /skill-tree/realms No List all 8 skill realms
GET /skill-tree/realms/{id} No Single realm with skill nodes
GET /skill-tree/progress Yes User's unlock/complete state + XP
POST /skill-tree/progress/{id}/unlock Yes Unlock a skill node
POST /skill-tree/progress/{id}/complete Yes Complete a skill & earn XP
GET /skill-tree/analytics Yes Learning analytics dashboard

Project Structure

HTE-DreamHackers-Submission/
├── frontend/                     React + TypeScript SPA
│   ├── src/
│   │   ├── pages/                Route components (14 pages)
│   │   ├── components/           UI components (65+ shadcn/ui + custom)
│   │   ├── contexts/             Auth state management
│   │   ├── hooks/                Custom React hooks
│   │   └── lib/                  API client, auth, utilities
│   └── package.json
├── backend/
│   ├── app/                      FastAPI REST API
│   │   ├── routers/              Endpoint handlers (auth, career, skill tree)
│   │   ├── schemas/              Pydantic request/response models
│   │   └── utils/                Auth helpers
│   ├── agents/                   LangGraph AI pipeline
│   │   ├── src/
│   │   │   ├── agents/           Pipeline nodes (NLP, vector, evaluator)
│   │   │   ├── data/             DB clients, embeddings, AI risk scoring
│   │   │   ├── outputs/          Learning path generation
│   │   │   ├── graph.py          Pipeline orchestration
│   │   │   ├── llm.py            LLM client with failover
│   │   │   └── config.py         Configuration & realm mappings
│   │   └── scripts/              Data pipeline scripts
│   ├── terraform/                AWS infrastructure (EC2, ECR)
│   ├── Dockerfile
│   └── pyproject.toml
└── README.md

Hackathon Track Submissions

1. Minimax Creative Usage Award

CareerQuest uses the Minimax API (MiniMax-Text-01) as the primary LLM powering the entire career advisor pipeline:

  • NLP Agent — Minimax parses natural language student input into structured career profiles (goals, skills, interests, constraints) using Pydantic schema enforcement.
  • Career Evaluator — Minimax reviews 10 semantically matched careers and selects the top 3 using a 5-dimension rubric (student fit, growth outlook, AI resilience, salary trajectory, accessibility).
  • Learning Path Generator — Minimax generates personalized 4-horizon roadmaps with pace modifiers (intensive/standard/relaxed), producing concrete action items from high school through post-graduation.
  • AI Risk Signal Evaluation — Minimax evaluates live market signals (research papers, news articles, job postings) to score career automation risk per domain.

The creative usage lies in chaining Minimax through a multi-agent state machine — each agent builds on the structured output of the previous one, producing a comprehensive career advisory response from a single free-text query. The fault-tolerant design with AWS Bedrock and OpenRouter as fallbacks ensures the pipeline never fails for the end user.

2. RevisionDojo

UI/UX & Engagement:

  • Clean, responsive interface built with shadcn/ui and Tailwind CSS, featuring dark/light mode toggle for user preference
  • Duolingo-style scrolling with visual progression indicators for attention retention
  • Gamification elements: XP system, 8 skill realms, hierarchical unlock trees, particle effects, sound effects, achievement badges, and level progression (Explorer → Legend)
  • 5-step guided onboarding with mascot messages and emoji-rich activity/interest selection

AI Pipeline Sophistication:

  • Multi-agent LangGraph pipeline incorporating NLP (structured extraction) and RAG (vector similarity over 1,000+ career embeddings)
  • Career evaluation uses Rubric-LLM prompting — a refined technique where the LLM scores candidates against an explicit 5-dimension rubric rather than generating free-form recommendations
  • Scalable LLM selection through fault-tolerant provider chain (Minimax → Bedrock → OpenRouter)
  • Token efficiency through vector database pre-filtering: 1,000+ careers → 10 vector matches → 3 LLM-evaluated selections, minimizing prompt size

3. OAX

Real-World Learner Impact: CareerQuest addresses a critical gap — high school students lack access to personalized, data-driven career guidance. Traditional career counseling is time-limited (one counselor per 400+ students), generic, and disconnected from real-time labor market data. CareerQuest makes institutional-quality career guidance accessible to any student with a browser.

Time-Saving & Learning Acceleration: What would take weeks of manual research — comparing careers, understanding AI automation risk, mapping educational pathways, finding relevant courses — is compressed into a single conversation. The multi-agent pipeline processes a natural language query and returns 3 evaluated career matches, each with a complete 4-horizon roadmap and matched courses, in under 30 seconds.

Personalization: Every recommendation is personalized through multiple layers:

  1. NLP extracts the student's unique profile (skills, interests, goals, constraints)
  2. Vector search finds careers semantically aligned to their profile, not just keyword matches
  3. Career evaluator considers accessibility — how realistic the path is given the student's current position
  4. Roadmaps adapt pacing (intensive/standard/relaxed) and content to the individual
  5. Course matching uses keyword overlap with the student's skill gaps

Creative Thinking: The combination of gamification (making career exploration feel like a game) with rigorous AI analysis (RAG, multimodal risk scoring, rubric-based evaluation) is novel. The AI automation risk scoring — blending structural labor data with live market signals — gives students forward-looking insight that no traditional career tool provides.

4. Ingram Micro & AWS

AWS Bedrock Integration: CareerQuest uses AWS Bedrock as the secondary LLM provider in a fault-tolerant pipeline:

Minimax API (Primary) ──▶ AWS Bedrock (Secondary) ──▶ OpenRouter (Fallback)

If the primary Minimax API experiences downtime, rate limiting, or errors, the pipeline automatically falls back to AWS Bedrock hosting Minimax models. This ensures zero-downtime for end users — the career advisor always responds, regardless of individual provider availability.

AWS Bedrock AgentCore: The platform leverages Bedrock AgentCore for production hardening:

  • AgentCore Evaluations — 13 built-in evaluators (correctness, helpfulness, safety) validate NLP parsing accuracy, career recommendation relevance, and roadmap quality
  • AgentCore Memory — Persists user context across sessions, enabling the platform to remember past career explorations and build preference profiles over time
  • AgentCore Policy — Natural-language guardrails enforce rules like "only recommend careers in the database" and "don't suggest careers requiring 10+ years experience to high school students"

AWS Infrastructure (Terraform):

  • EC2 — Hosts the FastAPI backend with Docker containerization
  • ECR — Container registry with lifecycle policies (retain last 5 images)
  • IAM — Instance roles with least-privilege access (ECR pull only)
  • Security Groups — Ingress restricted to HTTP (8000) and SSH (22)

Business Value: CareerQuest demonstrates how AWS services can power AI-driven education technology at scale. The combination of Bedrock (LLM inference), AgentCore (evaluation + safety), and EC2/ECR (deployment) provides a complete cloud-native stack for agentic AI applications. The fault-tolerant design ensures SLA compliance even during provider outages.

Innovation & Originality: The cascading LLM failover pattern (Minimax → Bedrock → OpenRouter) is a production-ready design pattern for multi-provider AI architectures. Rather than being locked into a single LLM vendor, the system dynamically routes requests based on availability, latency, and cost — a pattern applicable to any enterprise AI deployment.

5. HKUST Entrepreneurship Center

Innovation & Originality: CareerQuest introduces several novel elements to career guidance technology:

  • Multi-agent career pipeline — Rather than a single LLM call, the system uses a 5-stage pipeline where each agent specializes in one task (parsing, retrieval, evaluation, bridging, path generation), producing higher-quality results through specialization
  • AI automation risk scoring — A first-of-its-kind composite score blending structural labor data (O*NET, Frey-Osborne) with 5 live market signals, giving students forward-looking career intelligence
  • Gamified career exploration — Applying game mechanics (XP, skill trees, achievements) to career planning is novel and addresses the engagement gap that makes students abandon traditional career tools

Entrepreneurial Potential:

  • Large addressable market — 16.5M high school students in the US alone, plus international markets. Career guidance is a universal need with no dominant AI-native solution
  • B2B2C model — Can be sold to school districts, career counseling centers, and workforce development programs as a SaaS platform, while offering a freemium tier for individual students
  • Revenue streams — Premium features (unlimited advisor sessions, detailed analytics), institutional licensing, course platform partnerships (referral revenue from matched courses)
  • Network effects — More users → better career matching data → improved recommendations → more users
  • Low marginal cost — AI inference costs per query are minimal; the vector database and career data pipeline scale to millions of users without proportional cost increase

Technical Execution:

  • Full-stack implementation: React frontend + FastAPI backend + LangGraph AI pipeline + Supabase database + Terraform infrastructure
  • Production-ready CI/CD via GitHub Actions with Docker containerization
  • Type-safe throughout: TypeScript (frontend) + Pydantic v2 (backend) + structured LLM output
  • Security: JWT authentication, PII log redaction, Row-Level Security in database, CORS configuration

Real-World Impact: CareerQuest democratizes career guidance — making data-driven, AI-powered career planning available to students regardless of their school's resources. A student at an underfunded rural school gets the same quality of career intelligence as one at an elite prep school. The AI automation risk scoring helps an entire generation make future-proof career choices in an era of rapid technological change.


Testing

cd backend
uv pip install -e ".[test]"
pytest tests/ -v

Team

DreamHackers — Hack The Experience (HTE) 2026


License

This project was built for the HTE hackathon. All rights reserved.

Built With

Share this project:

Updates