AI-Powered Career Recommendation System Using Hybrid Architecture, Semantic Retrieval, and Dynamic Scoring

Abstract

The rapid growth of career options in a technology-driven job market has made intelligent and personalized career guidance an urgent necessity, particularly for students and early-career professionals who have limited access to professional counsellors. Most existing career recommendation systems rely on static rule-based logic and single-factor profile matching, resulting in recommendations that are neither adaptive nor sufficiently explainable. This paper presents an AI-powered career recommendation system that integrates resume analysis, semantic career matching, real-time job retrieval, and a Retrieval-Augmented Generation (RAG) chatbot within a scalable microservice architecture. A centralized scoring engine combines the outputs of all modules into a unified weighted score, ensuring that recommendations remain consistent, transparent, and adaptive across different user contexts. The system utilizes the all-MiniLM-L6-v2 sentence embedding model to compute semantic similarity between user profiles, career descriptions, and job descriptions, replacing traditional keyword-overlap techniques with contextual semantic matching. Design-based evaluation supported by controlled preliminary testing estimates career recommendation accuracy above 93%, module-level F1-scores of approximately 89% or higher, and user satisfaction close to 95%. With the adaptive feedback loop enabled, recommendation relevance is projected to improve by approximately 22 percentage points across repeated sessions. The proposed architecture compares favorably with IEEE-published baseline systems including Sankalp, CPRM, and the predictive advising model proposed by Hachaichi et al.