Hybrid (Rule-Based + Gen AI) Chatbot: Automating Negotiations

This paper proposes a hybrid framework combining rule-based systems with Large Language Models (LLMs) for automated supply chain negotiations. Building on successful industry implementations of rule-based negotiation bots and addressing vulnerabilities identified in MIT’s 2025 LLM negotiation competition, the approach leverages game-theoric principles while utilizing LLMs for natural language tasks. The natural language tasks include Constraint and Offer Interpreters, and Message Generation. In a nutshell: 1) Prompt the LLM with game-theoretic offers. 2) Check if the generated message is profitable enough. A rule-based wrapper ensures Pareto-efficient outcomes by validating LLM outputs post-generation before sending them to the counterpart. Although designed for procurement negotiations, researchers and practitioners can apply this framework to various negotiation contexts. It is entirely developed in Python, a widely-used open-source programming language.