Profile Knowledge Graph: An AI-Powered Resume Selection and Candidate Matching Platform

To develop and deploy an AI-powered resume selection and candidate matching platform to efficiently identify suitable candidates from unstructured resume data.

Traditional methods and legacy ATS are time-consuming, biased, and keyword-based, failing to capture semantic relationships. This project improves precision by 38-41% over TF-IDF keyword methods.

Automated resume parsing/candidate shortlisting for university placement cells, managing research lab projects (VICAS Hub), visual skill gap analysis, and complex multi-criteria talent discovery.

This report presents the exhaustive design, architectural methodology, implementation, and successful production deployment of a comprehensive AI-powered resume selection and candidate matching platform, delivered alongside a full-stack research lab manage- ment platform (VICAS Hub). In the contemporary recruitment landscape, the sheer volume of unstructured candidate data heavily burdens academic placement cells. To mitigate this, the system actively processes exactly 171 student resumes collected from professional networks such as LinkedIn, extracting structured, high-fidelity information using an optimized 5-stage Gemini 2.5 Flash API pipeline. The architecture leverages Facebook AI Similarity Search (FAISS) for sub-millisecond, high-dimensional vector re- trieval. It stores the semantic data in both MongoDB for traditional queries and Neo4j for highly visual, graph-based relationship exploration. A critical enhancement implemented in this final deployment phase is the Docker- ized Placement Cell Integration module, alongside our live cloud deployment. Built via FastAPI and React Vite, this containerized standalone architecture empowers placement officers to securely upload new resumes locally and automatically parse the entire 171- student cohort against complex corporate Job Descriptions (JDs) to instantly generate AI-reasoned, ranked shortlists. Concurrently, the VICAS Hub serves as a fully deployed 3-stage role access portal for managing lab research projects, moderated photo galleries, and features a sophisticated Academic BTP Portal driven by an automated Node-Cron mailing engine. Both systems have been rigorously refined, completed, and deployed to production environments, demonstrating a massive 38% increase in precision for complex multi-criteria queries over traditional TF-IDF keyword-based search methodologies.

=== Page 3 === Contents