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

Sep 11, 2026

Upload by Maeve

AI-assisted resume tailoring tools often rewrite only the current uploaded draft, limiting recovery of relevant experience stored in older resumes and structured career records. Resume Tailor is an agentic system that builds job-specific resume content through multisource retrieval-augmented generation from a longitudinal career vault. A provenance-aware pipeline separates retrieved, user-grounded text from fallback model outputs and applies anti-hallucination guardrails with a conditional review loop. Pilot results on nine job descriptions show average ATS-style gains when prior roles exist, and declines when domain overlap is weak.

LanguageEnglish

CategoryResumes

FormatPDF

Pages6

Reading Time2 min

AccessFree to download

Introduction

Motivation and design requirements

Related work

Resume-job matching

Explainable hiring pipelines

Retrieval-augmented generation

Agentic and iterative AI workflows

System architecture

Overview

Career-Aware Resume Tailoring via Multi-Source Retrieval-Augmented Generation with Provenance Tracking: A Case Study Kumar Abhinav [rushtoabhinavin@gmail.com](rushtoabhinavin@gmail.com) Abstract— AI-assisted resume tailoring systems commonly operate on a single uploaded resume, which limits their ability to recover relevant experience omitted from the current draft and makes it difficult for users to distinguish grounded edits from model-generated suggestions. This paper presents Resume Tailor, an agentic resume-tailoring system that maintains a longitudinal career vault in a vector database and uses multisource retrieval-augmented generation (RAG) to assemble job-specific resume contentfrom historical resumes and structured career records. The system is implemented as a 12-node LangGraph pipeline with typed state management, hybrid semantic-lexical confidence scoring, provenance-aware fallback generation, anti-hallucination guardrails, and a conditional review loop. We report a pilot evaluation on nine job descriptions (JDs) across software engineering, data analytics, and business analysis roles using a single candidate’s career history. For six JDs where the candidate held at least one prior role in the same occupational category, enabling the career vault improved Applicant Tracking System (ATS)-style fit scores by an average of 7.8 points. For two JDs requiring domain-specific expertise absent from the vault, scores decreased by an average of 8.0 points. One partially overlapping role showed a modest gain of 2 points. These results suggest that longitudinal retrieval can improve resume tailoring when relevant prior experience exists, while also highlighting the need for confidencegated retrieval when domain overlap is weak. Keywords—resume tailoring, retrieval-augmented generation, agentic AI, provenance tracking, ATS optimization, vector database, career history I. INTRODUCTION Applicant Tracking Systems (ATS) shape how resumes are screened before human review. In practice, candidates often adapt the same base resume across multiple applications, emphasizing different experiences depending on role, industry, and required skills. This process is time-consuming and errorprone, and it is increasingly supported by AI-assisted writing tools. Most current AI resume-tailoring workflows treat each request as a single-document task: the system rewrites only the uploaded resume for the current application. That assumption is restrictive because a candidate’s relevant evidence is often distributed across older resume versions, archived application materials, and structured career records. As a result, useful experience can remain inaccessible even when it is genuinely part of the candidate’s history. This limitation motivates three design requirements. First, a tailoring system should recover relevant experience across time rather than only from the active resume draft. Second, it should separate retrieved, user-grounded content from newly generated suggestions so that candidates can review outputs responsibly. Third, it should retain useful refinements across sessions rather than restarting from scratch each time. To address these requirements, we present Resume Tailor, a career-aware resume-tailoring system built around three ideas: (a) a longitudinal career vault implemented with persistent vector collections over historical resumes and structured career records; (b) provenance-aware generation in which retrieved content and fallback-generated content are tracked separately within the pipeline; and (c) a feedback loop that persists vetted generated content for reuse in later tailoring sessions. The system is implemented as a 12-node LangGraph pipeline with typed state, hybrid retrieval scoring, anti-hallucination guardrails, and a conditional holistic review step. We evaluate the approach in a pilot case study using nine real JDs and show where longitudinal retrieval helps, where it hurts, and why those

FAQs

Why do current AI resume-tailoring workflows underperform?

They typically treat each request as rewriting a single uploaded resume, making it hard to access relevant experience spread across historical resumes, archived materials, and structured career records.

How does Resume Tailor distinguish grounded content from model suggestions?

It uses provenance-aware generation to track retrieved content separately from fallback-generated content within the LangGraph pipeline, supported by anti-hallucination guardrails and a conditional review loop.

What were the pilot evaluation results across job descriptions?

When the candidate had at least one prior role in the same occupational category, the longitudinal career vault improved ATS-style fit scores by an average of 7.8 points. For roles lacking needed domain expertise in the vault, scores decreased by an average of 8.0 points, with one partially overlapping role showing a modest gain.

Sep 11, 2026

Upload by Maeve