An Intelligent Multi-modal Interview Simulation System Using Large Language Models, Automatic Speech Recognition, And Neural Text-to-speech Synthesis

Abstract

Preparing For Technical Employment Interviews Is A High-stakes Endeavor That Demands Both Domain Expertise And Practiced Verbal Communication. Conventional Preparation Strategies—textbook Study, Static Question Banks, And Peer Mock Sessions—suffer From Well-documented Limitations: They Are Non-personalised, Require Scheduling Coordination, And Provide No Systematic Feedback On Performance. This Paper Presents The AI Interview Assistant (AIIA), A Full-stack, Multi-modal Web Platform That Automates The Entire Interview Simulation Life-cycle. AIIA Integrates Three Distinct AI Services: (1) Google Gemini, A Large Language Model (LLM) Responsible For Context-aware Question Generation, Adaptive Conversational Follow-up, Code Evaluation, And Structured Feedback Synthesis; (2) Assem-blyAI Universal-2, A State-of-the-art Automatic Speech Recognition (ASR) Engine For Real-time Candidate Voice Transcription; And (3) Murf AI FALCON, A Neural Text-to-speech (TTS) Synthesiser That Voices The AI Interviewer Natalie. The System Supports Eight Technical Roles, Three Difficulty Tiers, And Three Code Chal-lenge Formats—write, Fix, And Explain—across Four Program-ming Languages. Interview Sessions Are Stored In A MongoDB Document Database, Enabling Longitudinal Progress Tracking. A Five-category, LLM-generated Feedback Report Is Delivered Upon Session Completion. Empirical Observations Demonstrate That The Five-prompt LLM Orchestration Architecture Produces Contextually Coherent Question Sets And Qualitatively Discrimi-native Performance Assessments. The AIIA System Establishes A Replicable Architectural Template For Deploying Conversational AI Agents In High-stakes Educational Assessment Contexts.

Keywords