The Analysis of BART-base and FLAN-T5 and Optimization using Optuna Optimizer for Improvement of Conversational AI to Enhance Customer Experience

This works looks into improving chatbot models for handling customer questions in specific fields. We focus on Tulippg.in, a site for finding PG housing information. We use BART-Base and FLAN-T5—two strong sequence models—trained on our own dataset. This set includes FAQs and user questions about housing. We tweak models with techniques like changing batch size, using gradient build-up, mixed precision (fp16), and Optuna optimization to make training smoother and answers better. To see how well these methods work, we use tests like BLEU, ROUGE, Accuracy, and BERT Score. Our tests show FLAN-T5 is good at making clear answers, while BART-Base can rephrase replies better. Both models get better with hyper parameter tuning using Optuna. This work shows that custom fine-tuning helps in boosting chatbot quality and offers a way to level up similar tools. This serves as a useful guide for making conversational AI work well in actual use.