AI Resume Analyzer using Generative AI

The problem is to analyse resumes and job descriptions consistently, identify evidence-backed skill and experience matches, and provide useful feedback while protecting applicant data and avoiding misleading or discriminatory recommendations. 1.2 Objectives Extract key resume sections such as education, skills, experience, and projects. Compare resume content with the requirements of a selected job description. Use semantic matching to recognise related skills beyond exact keyword overlap. Generate clear, evidence-based feedback and improvement suggestions. Evaluate output quality, privacy, fairness, and reliability. 1.3 Scope The system is intended as an academic decision-support prototype for resume review and self-improvement. It does not make hiring decisions, verify credentials, or guarantee employment outcomes. AI Resume Analyzer using Generative AI | Page 2

Resume screening has traditionally used keyword search, rule-based filters, and structured applicant-tracking systems. These approaches are relatively easy to interpret but may fail to recognise synonyms, context, or transferable skills. Natural language processing methods support text classification, named-entity extraction, and semantic representation, which can improve the handling of unstructured documents. Sentence embeddings and similarity measures can estimate the relationship between resume content and job requirements without requiring exact wording. Large language models introduce the ability to summarise experience and generate natural-language feedback. However, generated output may include unsupported details, inconsistent ratings, or bias. Consequently, a practical system should constrain prompts, preserve links to source text, and evaluate outputs against defined criteria rather than treating model-generated scores as objective truth.

A resume analyzer needs to combine semantic matching with transparent evidence and privacy-aware processing. A high similarity score alone cannot prove that a candidate is qualified. The proposed design addresses this gap by separating extraction, matching, and feedback generation, while making limitations visible to the user.

The proposed system accepts a resume file and a job description. It extracts text, identifies relevant sections, normalises skill terms, and builds semantic representations of resume statements and job requirements. A matching component estimates alignment and records supporting text. A generative AI component then converts these findings into a structured report containing matched skills, potentially missing requirements, unclear statements, and suggestions for improving clarity. The user can review and edit the report before using it.

Figure 1. Proposed processing flow (conceptual): Resume + Job Description → Text Extraction → Preprocessing → Skill/Experience Extraction → Semantic Matching → Evidence-Grounded Generative Feedback → Reviewable Report

A prototype may use Python for document processing, a PDF/DOCX parsing library for extraction, an embedding model for semantic representation, and a generative AI API or locally hosted language model for feedback. FastAPI can expose authenticated endpoints, while SQLite or PostgreSQL can store minimal report metadata. A web interface can allow users to upload a resume, paste a job description, inspect evidence, and download the report. Model and provider choices should be documented for reproducibility. AI Resume Analyzer using Generative AI | Page 3 4. Methodology

The prototype can be evaluated using consented sample resumes, synthetic examples, or appropriately licensed public data. Personal contact details should be removed or masked where they are not required. Uploaded resumes should not be used for model training without explicit permission, and retention periods should be clearly communicated. 4.2 Preprocessing Validate file type and size before parsing. Extract text and detect scanned documents that may require OCR. Separate resume sections such as skills, education, projects, and experience. Normalise common skill names while preserving the original wording. Record evidence snippets so that each reported match can be checked.

The system can combine exact skill matching with embedding-based semantic similarity. Job requirements should be represented as individual criteria rather than a single paragraph. Each criterion can be associated with a supporting resume excerpt or marked as not found in the supplied document. The language model should receive only the relevant extracted content and be instructed not to invent qualifications, infer protected characteristics, or treat missing information as proof that a person lacks a skill.

The report may include a summary of relevant experience, matched requirements with evidence, requirements that are not clearly demonstrated, resume clarity suggestions, and questions for the applicant to verify. If a numerical match indicator is included, its meaning and limitations should be explained; it should not be presented as a probability of job success.

1. Receive the resume and job description with user consent. 2. Validate files and extract text. 3. Segment resume and job requirements. 4. Extract and normalise skills and experience statements. 5. Calculate exact and semantic matches. 6. Attach evidence snippets to each finding. 7. Generate feedback using constrained prompts. 8. Validate the output and display it for human review.

Use authenticated access and restrict who can view uploaded documents. Encrypt data in transit and protect stored files. Keep API keys outside source code and use environment variables or a secret manager. Minimise personal data sent to external model providers. Provide clear retention and deletion controls. Log system errors without unnecessarily recording resume contents. AI Resume Analyzer using Generative AI | Page 4

7.1 Advantages

Recognises related terminology through semantic matching. Provides structured, readable feedback for applicants and reviewers. Links findings to evidence in the source resume. Can be extended to different roles and resume formats. 7.2 Limitations PDF parsing and OCR may fail on complex layouts or scanned files. Similarity scores can overstate or understate the actual relevance of a skill. Generative models can produce inaccurate or inconsistent feedback. Performance depends on model choice, prompts, language, and document quality. The system cannot independently verify the truth of a candidate's claims.

The system should be used as an assistive tool rather than an automated hiring authority. It should not infer sensitive personal characteristics or rank applicants based on protected attributes. Users should know how their information is processed, and applicants should be able to review and correct extracted information. Human reviewers must consider context and avoid treating a missing resume keyword as proof of missing ability.

Future work may include multilingual resume support, improved OCR, role-specific skill taxonomies, explainability views, local/private model deployment, integrations with applicant-tracking systems, and a larger evaluation with independent human reviewers. Further work should test reliability across industries, experience levels, and document formats. 10. Conclusion This paper proposed an AI Resume Analyzer using Generative AI to extract resume information, compare it with job requirements, and generate evidence-grounded improvement suggestions. The design combines document parsing, semantic matching, and constrained language-model feedback while emphasising privacy, transparency, fairness, and human review. Its practical usefulness must be established through systematic testing, and generated analysis should support—not replace—human judgement. AI Resume Analyzer using Generative AI | Page 5