An auditable no-code platform for building and evaluating hybrid conversational agents

Abstract

While Large Language Models (LLMs) have advanced conversational agents, creating reliable, governable chatbots remains a specialized task, often inaccessible to non-programmers. This paper introduces a hybrid, no-code platform that empowers domain experts to author, execute, and evaluate sophisticated chatbot conversations. The framework merges the control of deterministic decision trees with the flexibility of selective LLM integration. A visual, node-based editor allows authors to design conversation flows, which are exported as a strict, portable JSON schema. This schema decouples the design from a lightweight backend runtime that interprets the flow, executing deterministic steps and invoking LLMs only at designated points. The platform is domain-agnostic and incorporates privacy-by-design principles aligned with GDPR through explicit consent nodes and data minimization. A mixed-methods evaluation measured authoring usability via the System Usability Scale, end-user experience, and conversational effectiveness against rule-only and model-only baselines. The results demonstrate that the hybrid approach significantly lowers the barrier for non-technical authors, achieving higher task completion and efficiency than baseline models while ensuring authorial control over critical conversational steps. The primary contributions include a reusable schema, a practical runtime pattern for controlled LLM handoffs, and an evaluation toolkit, providing a scalable foundation for future research in no-code conversational AI.

Ähnliche Arbeiten