ELIPSE: Enhanced Live Interview Practice with Sentiment Evaluation, A Zero-Cost Real-Time AI Interview System Achieving Human Level Interaction Fidelity with Sub Two Second Latency and Multi-Dimensional Behavioral Analytics

Background:Interview preparation is among the final significant areas of workforce development that are resistant to automation, and the worldwide coaching sector is greater than $3 billion a year; however, more than 80 percent of job seekers in emerging economies do not have access to formal practice [1].With innovations in large language models (LLMs), neural text-to-speech (TTS), and transformer-based affective computing, artificially intelligent technology has created an unprecedented opportunity to democratize accessto high-quality interview simulation.Objective: ELIPSE (Enhanced Live Interview Practice with Sentiment Evaluation), to the best of our knowledge, is the first open-source platform to integrate, within a single system,real-time voice-to-voice interaction, adaptive LLM-driven question generation, multi-dimensional behavioral analytics (seven-class emotion detection, continuous sentiment scoring, hesitation quantification, speech-pace monitoring, and composite confidence estimation), and persona.Methods: ELIPSE is built on a React.js/Node.js WebSocket-driven architecture, integrating Groq-hosted LLaMA 3.3 70B inference, Microsoft Edge neural TTS, and a HuggingFace transformer frontend. Assessment used a within-sub longitudinal designwith 15 participants (47 completed sessions, 376 question-answer interactions) addressing three research questions related to latency, behavioral metric validity, and improvement in confidence.Results:ELIPSE achieves a mean end-to-end response latency of 1.83 s (SD = 0.41, 95% CI [1.79, 1.87]; 95th percentile= 2.61 s), a 26.4% reductionversus [2](3.8 s) and 42.8% lowerthan [3](3.2 s). Automated behavioral analytics correlated moderatelyto strongly with expert ratings (overall quality r = 0.71, 95%CI [0.62, 0.79], p