Career Genie: AI Career and Scholarship Guidance
Summary
My final-year project: a prototype that matches students to career paths and scholarships from a verified dataset, using RAG so each recommendation traces back to a real source.
My part
Career Genie: AI Career and Scholarship Guidance
My final-year project. It is a working prototype, not a finished product.
Executive Summary
Career Genie set out to address a gap in career counseling: most students receive generic advice that is not aligned with evolving job markets or verified scholarship opportunities.
The goal was a prototype that could:
- Generating personalized career paths
- Matching users with relevant scholarships
- Identifying skill gaps
- Providing resume feedback
- Delivering real-time conversational guidance
The system was designed using a Retrieval-Augmented Generation (RAG) architecture to ensure outputs are grounded in structured datasets rather than hallucinated responses.
The Problem
Traditional career counseling suffers from:
- Static and outdated guidance
- Fragmented scholarship information
- Lack of personalization
- No explainability in recommendations
- Limited access for students in emerging markets
Most platforms either:
- Provide assessments only
- Offer scholarship listings only
- Or rely purely on LLM chat without grounded data
There was no unified intelligence layer combining career data, scholarships, and AI reasoning into a single system.
The Solution
Career Genie integrates:
- Structured career datasets (O*NET-based)
- Verified scholarship data (government + international sources)
- Semantic search via vector embeddings
- LLM reasoning for explanation and personalization
The result is a contextual, explainable AI system that delivers:
- 3–5 best-fit career paths
- Eligibility-matched scholarships
- Course recommendations based on skill gaps
- Resume analysis aligned to selected career paths
- A real-time AI chatbot grounded in retrieved knowledge
System Architecture
1. Data Layer
- Career dataset collected and normalized from O*NET
- Scholarship dataset curated via scraping and manual validation
- Structured storage in PostgreSQL
- Embedding generation using
gemini-embedding-001
2. Retrieval Layer
- Vector embeddings stored in Supabase (pgvector)
- Semantic similarity search for:
- Career alignment
- Scholarship eligibility
- Skill-to-role mapping
3. Intelligence Layer (RAG)
- Retrieved context injected into prompts
- Gemini 2.5 Flash generates:
- Personalized recommendations
- Resume feedback
- Assessment explanation
- Chat responses
This prevents hallucinations and ensures grounded outputs.
4. Application Layer
- FastAPI backend for orchestration
- Next.js frontend for dashboards and onboarding
- Streaming chatbot interface (SSE-based)
- Role-based views for students and institutes
Engineering Decisions
Why RAG Instead of Pure LLM?
To avoid hallucinated career or scholarship data and ensure explainable outputs grounded in verified datasets.
Why Gemini 2.5 Flash?
Balanced reasoning capability and latency efficiency for interactive applications.
Why Supabase + PostgreSQL?
- Native pgvector support
- Structured + vector hybrid queries
- Rapid deployment and scalability
How the Prototype Performed
- Career recommendation generation: 20–30 seconds
- Resume analysis: 10–15 seconds
- Chatbot response latency: 3–5 seconds
- Functional testing: 15/15 test cases passed
Outcome
Career Genie showed that grounding recommendations in structured career and scholarship data, instead of letting a model answer freely, gives answers you can trace back to a source.
.jpg%3Ftoken%3DeyJraWQiOiI1ZDU2MzVjZC1lZDU0LTQ1ZDItYmFjOC1lMmYyNGJmZTY3NTkiLCJhbGciOiJIUzI1NiJ9.eyJ1cmwiOiJwcm9qZWN0cy9nYWxsZXJ5LzJkZjQ1Yjk5LWQ1ZjMtNDBkNi1iMzg5LTMwNTUxMTM4ZWU0Ni8xNzcxMDUyNzIzNTMzLU1PQ0tVUCA1ICgxKS5qcGciLCJzY29wZSI6ImRvd25sb2FkIiwiaWF0IjoxNzk0OTczMDEyLCJleHAiOjE3OTQ5ODAyMTJ9.T0v843y95ShVnxkpPgenvQ2lZ7LEVaaN8Whz7E1C1iE&w=3840&q=75)
System Visuals
.jpg%3Ftoken%3DeyJraWQiOiI1ZDU2MzVjZC1lZDU0LTQ1ZDItYmFjOC1lMmYyNGJmZTY3NTkiLCJhbGciOiJIUzI1NiJ9.eyJ1cmwiOiJwcm9qZWN0cy9nYWxsZXJ5LzJkZjQ1Yjk5LWQ1ZjMtNDBkNi1iMzg5LTMwNTUxMTM4ZWU0Ni8xNzcxMDUyNzIzNTMzLU1PQ0tVUCA1ICgxKS5qcGciLCJzY29wZSI6ImRvd25sb2FkIiwiaWF0IjoxNzk0OTczMDEyLCJleHAiOjE3OTQ5ODAyMTJ9.T0v843y95ShVnxkpPgenvQ2lZ7LEVaaN8Whz7E1C1iE&w=3840&q=75)
Have a similar engineering problem?
Let’s discuss your requirements, constraints, and what a useful solution would look like.