Engineering a Privacy-First, Locally Executable AI Interview Coaching System with Evidence-Traced Coaching and Consent-Gated Inference
Year: 2026
Type: preprint
Abstract: This paper presents MindOne, a privacy-first, locally executable AI interview coaching system designed to provide response-specific, evidence-traced coaching while keeping candidate information local by default. The system converts résumé and optional job-description information into a structured interview plan using a compositional content matrix of 504 cells, covering 12 professional roles, 6 question types, and 7 difficulty levels. It provides turn-managed mock interviews, response-specific feedback, practice debriefs, recurring-gap identification, and competency guidance. A deterministic evaluation layer separates observed evidence, inference, and missing information. Numeric scoring is disabled by default. Remote AI inference is available only through explicit, versioned, and revocable consent. The paper describes the system architecture, evidence-first development methodology, and engineering verification. Verification includes contract and unit testing, socket-blocked network-egress tests, provider-failure injection, bounded-queue backpressure tests, and deletion probes. The manuscript also reports preliminary informal field use involving more than 50 college students across five colleges, together with additional voluntary candidates. This field use was not a controlled effectiveness study. No learning-gain, interview-success, employability, or anxiety-reduction claims are made. The work is presented as an engineering and system-design contribution. Its limitations, verification boundaries, and future independent evaluation requirements are explicitly documented. Version: 1.0.Author: Chandra Hotha, Independent Researcher, India.ORCID: https://orcid.org/0009-0008-6353-0988
Author: Hotha, Chandra
Language: English
Topic: AI and HR Technologies
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7226596041
