Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM69143
Volume:
10 |
11 |
Nov 2026
Pages:
1-12
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Department of Computer Science
MIT-ADT University
Department of Computer Science
MIT-ADT University
Department of Computer Science
MIT-ADT University
Department of Computer Science
MIT-ADT University
Department of Computer Science
MIT-ADT University
TRESK-AI is an implemented full-stack platform for AI-driven mock interview practice; this paper reports its architecture, adaptive interviewing mechanism, and evaluation methodology as they exist in the deployed system rather than as originally proposed. Because the platform depends on third-party, largely free-tier large language model (LLM) endpoints with no uptime or rate-limit guarantees, its most consequen-tial engineering contribution is an application-level reliability architecture – ordered multi-model failover through a gateway, a secondary direct-API fallback, response caching, and a fully deterministic rule-based fallback – that keeps question generation and answer evaluation available even when every upstream LLM call fails. Building on this foundation, the platform generates interview questions conditioned on a candidate’s resume and target role, and adjusts follow-up question depth using a thresh-old on a documented, per-answer evaluation score that blends an LLM-judged holistic rating with a deterministic, transcript-derived heuristic. Feedback is expressed along multiple named dimensions rather than a single opaque number. The platform additionally supports voice-based interaction through browser-native speech recognition with a cloud speech-to-text fallback, an optional sandboxed coding-assessment module built on the Piston execution engine, and lightweight anti-cheating monitors based on window-focus tracking. During implementation review, we determined that the platform’s on-screen gaze and stress indi-cators are client-side simulations rather than outputs of a genuine video-analysis pipeline; we report this finding directly and scope our claims accordingly (Section VI-E). We describe the system’s technology stack, its resume-conditioned question-generation and evaluation mechanisms, and a concrete experimental protocol – covering adaptive-versus-static question generation, resume-conditioned-versus-generic questioning, and AI-versus-human-rater evaluation agreement – designed to be executed on this implementation. As a first, informal step toward that evaluation, we additionally report preliminary code-level findings executed as part of this revision: measured automated-test coverage and a dependency-vulnerability scan. No controlled human-subjects study has yet been conducted, so we do not report accuracy, correlation, or usability figures from that protocol; we instead specify the metrics, baselines, and threats to validity relevant to interpreting such a study once it is run.
KEYWORDS
This paper re-examined TRESK-AI, a working mock-interview platform, directly against its source code rather than against its own documentation, and revised the platform’s description accordingly. The verified implementation includes resume-conditioned question generation, a threshold-gated follow-up adaptivity mechanism, a disclosed blended scoring formula combining an LLM judgment with a transcript-derived heuristic, a three-tier LLM reliability cascade now bounded quantitatively for its gateway stage (Eq. (2)), browser- and cloud-based voice interaction, a Piston-backed coding sand-box, and a real focus-tracking anti-cheating mechanism. The same review found that the platform’s webcam-based gaze and stress indicators are a client-side simulation with no underlying analysis pipeline (Section VI-E), and identified, as new find-ings specific to this revision, an untested failover module and two core AI-service modules with no automated test coverage at all (Section VI-B), a frontend dependency tree with two fixable moderate-severity advisories (Section VIII-A), and an unenforced timeout on the Gemini fallback path (Section III). We view the combination of a corrected implementation de-scription, a formalized (and honestly scoped) adaptive mech-anism, a small set of executed preliminary findings, and a concrete, not-yet-executed experimental protocol as a more useful contribution than a more polished description that could not be defended against the source code, and we identify executing that protocol as the immediate next step for this work.
[1] M. I. Tanveer, D. Gildea, and M. E. Hoque, “Automated analysis and prediction of job interview performance,” IEEE Transactions on Affective Computing, vol. 9, no. 2, 2018. [2] L. Chen, R. Zhao, C. W. Leong, B. Lehman, G. Feng, and M. E. Hoque, “Automated video interview judgment on a large-sized corpus collected online,” in Proc. IEEE Int. Conf. Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW), 2017. [3] L. K. Joshila Grace et al., “Automated interview through online video interface,” in Proc. IEEE Conference, 2023. [4] A. A. J. Karim, S. Hoque, M. G. R. Alam, and M. Z. Uddin, “Automated personnel selection for software engineers using LLM-based profile evaluation,” arXiv:2410.23365, 2024. [5] S. Hoque, A. A. J. Karim, M. G. R. Alam, and N. Gope, “When LLM meets Fuzzy-TOPSIS for personnel selection through automated profile analysis,” arXiv:2601.22433, 2026. [6] A. Fabris, N. Baranowska, M. J. Dennis, D. Graus, P. Hacker, J. Saldivar, F. Zuiderveen Borgesius, and A. J. Biega, “Fairness and bias in algorithmic hiring: A multidisciplinary survey,” ACM Transactions on Intelligent Systems and Technology, 2024. [7] L. Zheng et al., “Judging LLM-as-a-judge with MT-Bench and Chat-bot Arena,” in Advances in Neural Information Processing Systems (NeurIPS), 2023. [8] A. Szymanski, N. Ziems, H. A. Eicher-Miller, T. J.-J. Li, M. Jiang, and R. A. Metoyer, “Limitations of the LLM-as-a-judge approach for evaluating LLM outputs in expert knowledge tasks,” arXiv:2410.20266, 2024. [9] S. Vatsal and H. Dubey, “A survey of prompt engineering methods in large language models for different NLP tasks,” arXiv:2407.12994, 2024. [10] A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervision,” arXiv:2212.04356, 2022.
Publication Date
11/18/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
0
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...