Template-Driven Specification of Requirements for LLM-Based Chatbots

The integration of large language models (LLMs) into chatbots introduces specific requirements related to reasoning, regulatory constraints, and user-sensitive interactions. This paper presents a template-driven approach for specifying system requirements in LLM-enhanced chatbots. The method uses a structured requirements card that defines the functional and non-functional roles, input-output behaviors, contextual reasoning dependencies, and compliance or operational constraints. By formalizing these requirements, the approach supports alignment between LLM capabilities and system design, facilitates verification of behavior across dialogue flows, and improves traceability within the software architecture. This specification model is particularly suited for domains where consistency, interpretability, and compliance with business rules are critical. As a proof of concept, a set of requirements for a financial chatbot was elicited and organized into cards derived from the template.