This paper presents the design, implementation, and evaluation of a domain-specific intelligent chatbot built using the Rasa open-source framework, augmented with enhanced Natural Language Understanding (NLU) techniques. Modern conversational agents face significant challenges in accurately interpreting domain-constrained user queries — from intent ambiguity to multi-entity extraction in complex utterances. This work leverages Rasa's Dual Intent and Entity Transformer (DIET) classifier 10 alongside the Transformer Embedding Dialogue (TED) Policy to achieve high-performance intent classification and contextual dialogue management. The proposed system was evaluated on a healthcare/customer support domain dataset, achieving an intent detection accuracy of 92.1% and an entity recognition F1-score of 0.88. Experimental results validate the superiority of the context-aware pipeline over single-turn baseline NLU models, demonstrating a 25% improvement in human-centric evaluation metrics. The system demonstrates that combining Rasa's modular NLU pipeline with transformer-based embeddings enables scalable, interpretable, and accurate domain-specific conversational AI.6
Chandra et al. (Mon,) studied this question.