Skip to content
All projects
Applied AI ResearchDeployed · Punjab College of ExcellenceFeb 2025 – Jan 2026

Career Connect

AI-Driven Psychometric Career Guidance Platform

Career Connect is an intelligent career guidance and psychometric recommendation platform engineered to map student profiles against the comprehensive O*NET occupational database. It combines deterministic vector-space matching with a retrieval-augmented explanation layer, then packages the result as a production web application used by Punjab College of Excellence.

PythonFastAPINext.jsVector Space ModelRAGDockerMongoDB
Next.js client
FastAPI gateway
VSM matcher
RAG / O*NET
LLM explanation
MongoDB
FIG. 01Career Connect · inference flow

Challenge

Students receive generic career advice that is difficult to trust or act on. The system had to turn noisy interest, personality, and skill signals into ranked, explainable career paths — without letting a language model invent qualifications, and without requiring the institution to manage a research notebook. It also had to ship as a product that non-technical counsellors could use.

Architecture

Next.js captures assessment signals on the client. FastAPI validates and normalizes the profile, then runs two complementary paths: a deterministic VSM matcher applies weighted cosine similarity across multi-dimensional skill and interest vectors against O*NET, while a RAG pipeline retrieves occupational context to ground the LLM's explanations. MongoDB persists profiles, matches, and generated roadmaps. Docker packages the API and model services for reproducible deployment.

Key decisions

Deterministic matching before generation

VSM produces reproducible scores; the LLM is used to explain the match, not to invent it. This keeps rankings auditable and reduces hallucination.

O*NET as the occupational ground truth

Using a standard occupational taxonomy makes vectors comparable, recommendations defensible, and the system easier to evaluate.

Separate retrieval from generation

The RAG layer retrieves relevant occupational evidence first; generation synthesizes that evidence into a career roadmap with traceable context.

Containerized production deployment

Docker keeps the API, model dependencies, and runtime consistent from local development through the institutional deployment.

Results

  • Deployed to Punjab College of Excellence
  • Deterministic VSM ranking with grounded explanations
  • End-to-end product: Next.js, FastAPI, MongoDB, Docker
  • Production web interface for non-technical counsellors

© 2026 Muhammad Ali · alisardar.dev

Islamabad, Pakistan · Last updated November 2026