Career-Aware Resume Tailoring via Multi-Source Retrieval-Augmented Generation with Provenance Tracking
This academic paper introduces Resume Tailor, an agentic AI system designed to enhance resume customization by leveraging a longitudinal career vault. Unlike traditional systems that rely on a single uploaded document, this solution utilizes multi-source retrieval-augmented generation (RAG) to retrieve relevant experience from historical resumes and structured career records stored in a vector database. Implemented as a 12-node LangGraph pipeline, the system features typed state management, hybrid confidence scoring, and anti-hallucination guardrails to ensure grounded edits. A pilot evaluation involving nine job descriptions across software engineering, data analytics, and business analysis roles demonstrated its efficacy. For positions matching the candidate's prior occupational categories, the system improved Applicant Tracking System (ATS) fit scores by an average of 7.8 points. However, scores decreased when domain-specific expertise was absent from the vault, highlighting the necessity for confidence-gated retrieval mechanisms. The study underscores the potential of longitudinal data retrieval in AI-assisted career tools while identifying limitations in handling weak domain overlaps.
Wire timeline
Career-Aware Resume Tailoring via Multi-Source Retrieval-Augmented Generation with Provenance Tracking
This academic paper introduces Resume Tailor, an agentic AI system designed to enhance resume customization by leveraging a longitudinal career vault. Unlike traditional systems that rely on a single uploaded document, this solution utilizes multi-source retrieval-augmented generation (RAG) to retrieve relevant experience from historical resumes and structured career records stored in a vector database. Implemented as a 12-node LangGraph pipeline, the system features typed state management, hybrid confidence scoring, and anti-hallucination guardrails to ensure grounded edits. A pilot evaluation involving nine job descriptions across software engineering, data analytics, and business analysis roles demonstrated its efficacy. For positions matching the candidate's prior occupational categories, the system improved Applicant Tracking System (ATS) fit scores by an average of 7.8 points. However, scores decreased when domain-specific expertise was absent from the vault, highlighting the necessity for confidence-gated retrieval mechanisms. The study underscores the potential of longitudinal data retrieval in AI-assisted career tools while identifying limitations in handling weak domain overlaps.
cs.AI updates on arXiv.org