Authors:
Aashutosh Indusekhar, R. Manikandan, T. Sanjay Karthik, V. Nivedita, Targyn A. Nauryz, S. J. Vimal Aravintha

Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. School of Digital Technologies, Narxoz University, Almaty, Kazakhstan. Department of Data Science, Indiana University Bloomington, Bloomington, Indiana, United States of America.

Abstract:

The large number of applications that recruiters need to go through has made the limitations of traditional Applicant Tracking Systems (ATS) increasingly hard to ignore. The keyword-matching logic, lack of transparency in scoring, and vulnerability to resume cheating are issues that traditional systems have not addressed. This paper presents Resume AI, a hybrid agentic resume ranking system that uses a two-layer architecture. The first layer is a deterministic NLP engine that deals with all scoring tasks using a four-component weighted formula: Skill-set intersection, TF-IDF job description similarity, experience weighting, and a bounded LLM decision score. The second layer is the locally running large language model that interprets the structured resume and structured job description, makes candidate ranking decisions with a capped contribution of 20% to the overall score, and generates recruiter explanations. All this inference happens on the device with Ollama, with no data sent to any service. Tested with the keyword-only baseline on 15 resume-JD pairs, the system can correctly penalise modified resumes, handle semantic synonyms that exact matching would miss, and generate explanations based entirely on deterministic data for recruiters.

Keywords: Resume Parsing, Explainable AI, Hybrid AI Architecture, Agentic LLM, Job Description Matching, Skill Gap Analysis, Candidate Ranking, Recruitment Intelligence, Local LLM, Weighted Scoring, LLM Decision Agent.

Received on: 16/05/2025, Revised on: 11/07/2025, Accepted on: 12/10/2025, Published on: 12/06/2026

DOI: 10.69888/FTSCS.2026.000688

FMDB Transactions on Sustainable Computing Systems, 2026 Vol. 4 No. 2, Pages: 149-164

• Views : 46
• Downloads : 6

Download PDF