Career-Aware Resume Tailoring via Multi-Source Retrieval-Augmented Generation with Provenance Tracking: A Case Study

Resume Tailor tackles a key limitation of current AI resume tools: they only see one uploaded resume and can't recover omitted experience or distinguish grounded edits from hallucinated suggestions. The system stores a candidate's full career history in a vector database as a longitudinal vault. When a new job description arrives, it uses multi-source retrieval-augmented generation to pull relevant past roles and achievements from that vault, then generates a tailored resume with provenance tracking—clearly marking which edits come from real experience vs. model inference. The pipeline is built on LangGraph with 12 typed nodes, hybrid confidence scoring (semantic + lexical), a conditional review loop for human oversight, and anti-hallucination guardrails. This design lets the system recover experience the user may have omitted from their current draft and provides transparency into why each suggestion was made.

In a pilot study across nine job descriptions in software engineering, data analytics, and business analysis, Resume Tailor showed clear benefits when relevant experience existed. For six JDs where the candidate had at least one prior role in the same occupational category, enabling the career vault improved ATS-style fit scores by an average of 7.8 points. However, for two JDs requiring domain-specific expertise absent from the vault, scores dropped by an average of 8.0 points. One partially overlapping role saw a modest gain of 2 points. These results suggest that longitudinal retrieval can meaningfully improve resume tailoring when relevant experience exists, but also underscore the need for confidence-gated retrieval to prevent irrelevant content from degrading quality. The paper calls out that future systems should balance retrieval strength with domain overlap confidence.