Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{203494,
        author = {Sairaj Fattesing Jadhav and Mr Nitin Magdum},
        title = {Adaptive Technical Interview Preparation via LLM-Guided Question Regeneration, Section-Weighted Resume Parsing, and Dual-Path Speech–Text Evaluation},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {11329-11334},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=203494},
        abstract = {Technical interview preparation platforms commonly distribute static question pools disconnected from a candidate’s demonstrated skill profile, producing assessments that miss role-relevant depth. This paper presents an AI-driven web application addressing this gap through three contributions: a section-aware skill extraction algorithm, a large language model query chain that generates and adaptively regenerates interview questions between conversational turns, and a multi-dimensional evaluation engine scoring confidence, technical correctness, and communication clarity independently. The extraction algorithm assigns differential weights 3.0× for dedicated skills sections, 1.5× for experience and project entries, and 1.0× for unstructured body text with contextual phrase bonuses capped at 6.0 points per skill. After each candidate response, Google Gemini 2.5 Flash regenerates the subsequent question by conditioning on the full conversation history, producing continuity absent from static question banks. Voice answers streamed over an Assembly AI Universal-Streaming WebSocket at 16 kHz PCM-16 merge with typed input via an overlap-aware combination algorithm before evaluation. Against a manually annotated corpus of 200 résumés, section-weighted extraction achieves an F1 score of 0.777 compared with 0.673 for flat keyword matching and 0.700 for TF-IDF approaches. The platform eliminates question-preparation overhead while providing candidates with personalized, progressive interview practice grounded in their resume evidence.},
        keywords = {adaptive question generation, answer evaluation, interview preparation, large language models, resume parsing, speech transcription},
        month = {May},
        }