An Intelligent Skill and Job Recommendation System With Predictive Career Analytics

Job seekers manage applications, resumes, and outcomes across fragmented tools, reducing visibility into progress and making it difficult to quantify job fit or prioritize skill development. This paper presents a Job AI System that unifies application tracking with resume intelligence, semantic retrieval, and decision support. The platform integrates PostgreSQL for transactional data with ChromaDB for persistent vector retrieval over resumes and job postings. Resume ingestion supports PDF/DOCX parsing and structured extraction (skills, education, experience, summary) via a fallback-first pipeline, invoking an LLM when deterministic parsing is insufficient. Job–resume alignment is computed using a hybrid scoring model that combines dense embedding similarity (Sentence-Transformers all-mpnet-base-v2), skill overlap and semantic skill matching, TF–IDF-weighted signals, and experience/education heuristics, with optional LLM-based refinement for borderline cases. The system also provides deterministic outcome estimation (interview and offer probabilities) with caching and a “what-if” simulation module that estimates marginal gains from adding hypothetical skills. An AI Copilot interface enables retrieval-augmented generation over user data using an LLM via Ollama with server-sent event streaming. The proposed design demonstrates an end-to-end, deployable architecture for datadriven job search guidance using hybrid retrieval, matching, and simulation.