D16ADB Group 17 (2025-26) Hybrid AI Interview Coach

The fast pace of development in generative AI has completely changed the field of education technology, yet the realm of interviewing preparation software is largely comprised of static, text-driven tools ignoring the communicative value of oral communication. This paper highlights the limitations of existing interviewing practice approaches, such as preparing to answer questions from a bank and practicing alone, in creating the essential real-time feedback loop necessary to master speaking skills in technical environments. In response, a new hybrid interview coach is proposed, using Retrieval-Augmented Generation (RAG) to adjust content according to resumes, while employing real-time Digital Signal Processing (DSP) technology for voice evaluation. Unlike the unpredictable autonomous generation agent, a hybrid model with adaptive behavior con trolled via state heuristics is used. The proposed hybrid system architecture consists of a React front-end interface, adaptive Flask back-end controller utilizing concept mastery velocity calculations, semantic duplicate filter, Sentence-BERT+FAISS RAG pipeline, Mistral API for generation, AssemblyAI for live transcript extraction, and YIN framework for estimating stream ing pitch stability, pause ratio, and WPM rate. This paper outlines architectural choices, algorithms, potential limitations, and possible future research directions in developing the system of multimodal interviewing coaching.

Видео D16ADB Group 17 (2025-26) Hybrid AI Interview Coach канала VESIT AI&DS