AI-Powered Mock Interview Platform for Personalized Multimodal Interview Preparation and Performance Evaluation

Abstract Modern campus placement and technical recruitment pipelines demand a multilayer evaluation of graduating software engineers, encompassing algorithmic problem-solving proficiency, structured verbal reasoning, and composed nonverbal delivery. However, scalable mock interview preparation remains constrained by human-resource-intensive faculty panels and the summative, non-pedagogical design of commercial video screening software. This paper presents an integrated, privacy-preserving multimodal mock interview platform tailored to formative placement training and objective competency evaluation. The decoupled multi-tier architecture combines generative language model reasoning (LLaMA 3.3 70B via Groq Language Processing Unit acceleration), client-side facial and iris landmark telemetry using Google MediaPipe Face Landmarker in WebAssembly, browser-native acoustic speech transcription, neural vocal synthesis via ElevenLabs, and an isolated Wandbox Docker compiler sandbox integrated with a Monaco code editor. Because computer vision inference executes exclusively within volatile client-side browser memory at 5 frames per second (FPS) before immediate frame disposal, the platform requires 0.0 kbps of cloud video streaming bandwidth and preserves candidate biometric privacy with sub-12 ms local tracking latency. A normalized late-fusion assessment algorithm synthesizes verbal semantic depth (55%), vocal cadence and articulation (25%), and visual gaze stability (20%) into auditable multi-tier diagnostic scorecards. Empirical validation across 39 multi-stage simulation sessions (1,367 cumulative practice minutes) achieved 94.2% precision in structured behavioral evaluation, 92.0% steady gaze retention, a 100% test-case pass rate across sandboxed programming challenges, an Applicant Tracking System (ATS) readiness score of 85/100, and an average longitudinal competency gain of +26.4% alongside a reduction in speech disfluencies from 4.2 to 0.2 fillers per minute.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11485890/v1
Primary Topic
AI and HR Technologies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

AI-Powered Mock Interview Platform for Personalized Multimodal Interview Preparation and Performance Evaluation

Sitheswar K, Siva M, R. Vinoth Kumar
2 links
Research Square
AI and HR Technologies
preprint

AI-Powered Mock Interview Platform for Personalized Multimodal Interview Preparation and Performance Evaluation

Sitheswar K, Siva M, R. Vinoth Kumar
preprint en

Abstract

Abstract Modern campus placement and technical recruitment pipelines demand a multilayer evaluation of graduating software engineers, encompassing algorithmic problem-solving proficiency, structured verbal reasoning, and composed nonverbal delivery. However, scalable mock interview preparation remains constrained by human-resource-intensive faculty panels and the summative, non-pedagogical design of commercial video screening software. This paper presents an integrated, privacy-preserving multimodal mock interview platform tailored to formative placement training and objective competency evaluation. The decoupled multi-tier architecture combines generative language model reasoning (LLaMA 3.3 70B via Groq Language Processing Unit acceleration), client-side facial and iris landmark telemetry using Google MediaPipe Face Landmarker in WebAssembly, browser-native acoustic speech transcription, neural vocal synthesis via ElevenLabs, and an isolated Wandbox Docker compiler sandbox integrated with a Monaco code editor. Because computer vision inference executes exclusively within volatile client-side browser memory at 5 frames per second (FPS) before immediate frame disposal, the platform requires 0.0 kbps of cloud video streaming bandwidth and preserves candidate biometric privacy with sub-12 ms local tracking latency. A normalized late-fusion assessment algorithm synthesizes verbal semantic depth (55%), vocal cadence and articulation (25%), and visual gaze stability (20%) into auditable multi-tier diagnostic scorecards. Empirical validation across 39 multi-stage simulation sessions (1,367 cumulative practice minutes) achieved 94.2% precision in structured behavioral evaluation, 92.0% steady gaze retention, a 100% test-case pass rate across sandboxed programming challenges, an Applicant Tracking System (ATS) readiness score of 85/100, and an average longitudinal competency gain of +26.4% alongside a reduction in speech disfluencies from 4.2 to 0.2 fillers per minute.

Research Square
Rajalakshmi Engineering College (IN)
Quality education
AI and HR Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon