Automated Resume Screening System

- Many automated applicant-screening tools depend on word overlap, so they cannot see that two differently phrased resumes may describe the same expertise, and they return scores that neither recruiters nor candidates are able to examine. This paper presents an open-source, explainable hybrid framework for resume screening that runs on ordinary CPUs. Uploaded PDF, DOCX or plain-text files are cleaned and parsed, extracted skills are converted to standard labels, and a lightweight sentence encoder places the resume and the job posting in one shared 384-dimensional vector space. The overall job-fit value blends semantic similarity, skill coverage and an experience adjustment, and an extended profile also takes project and education relevance into account. Each matched skill is linked to a supporting sentence in the resume, a presentation-quality index is computed separately from job fit, rule-based integrity checks flag questionable content, and a vector database supports semantic retrieval and near-duplicate detection. The paper states the scoring model and the processing algorithm, outlines the relational and vector storage layers, and summarizes the functional checks carried out in an implementation audit. The audit also exposed open problems, chiefly fixed weights, integrity results that are not stored, missing access control and the lack of a dispute mechanism; these are reported openly, together with a reproducible protocol for assessing ranking quality