Contextual Embedding-Based Resume Screening with Semantic Relevance Scoring and Qualified Candidate Retention

Modern hiring platforms face mounting pressure to process large volumes of applications without sacrificing candidate quality. Conventional screening tools built on token- matching logic are structurally incapable of recognizing con- ceptual alignment between applicant profiles and position re- quirements, resulting in disproportionate rejection of genuinely qualified individuals. This study presents an intelligent recruitment screening system that combines deep contextual text representations, high-speed vector retrieval, and structured competency mapping. A purpose- built relevance annotation workflow is embedded directly into the hiring pipeline to support credible system benchmarking. The platform is constructed around a service-oriented architecture using FastAPI for server-side logic, React for the recruiter interface, and FAISS for vector-based candidate retrieval. Controlled experiments confirm that the proposed approach substantially outperforms both lexical and static-embedding base- lines on ranking quality, retrieval coverage, and candidate reten- tion metrics. Query response times remain below 100 millisec- onds, confirming viability for production-scale deployment. The findings underscore the value of deep language understanding in building equitable and efficient AI-assisted hiring tools.