Conditional Text Generation for AI ‐Powered Interviews: A T5 ‐Based System With GPT ‐2 Comparison

Artificial intelligence (AI) has transformed traditional processes by enhancing scalability, accuracy, and efficiency across industries. One area ripe for reform is the interview process. This paper presents an AI‐powered interview and assessment system that leverages natural language processing to evaluate candidates in real time, aiming to promote impartial and consistent evaluations. At the core is a T5‐based conditional text generation model trained to generate MERN stack–related technical questions from a dataset of 5462 questions, prepared through web scraping and synonym‐based data augmentation. Experimental results show that T5 significantly outperforms a GPT‐2 baseline across multiple metrics: BLEU (34.96 vs. 19.21), ROUGE‐1 (0.58 vs. 0.20), ROUGE‐2 (0.43 vs. 0.23), ROUGE‐L (0.56 vs. 0.34), METEOR (0.50 vs. 0.35), Perplexity (1.07 vs. 1.90), and Cosine Similarity (0.73 vs. 0.62). These improvements highlight T5's superior ability to generate coherent, contextually relevant questions, underscoring the potential of AI‐driven interview systems to make recruitment more scalable, fair, and effective.