Reliable AI Agents

AI agent demos based on LLMs are so common, but as Tinkerers we know they have many pitfalls around accuracy and hallucination. If we alter the prompt just slightly, the output completely changes! Grounded in years of dialogue systems research, I will be presenting the latest features in an agent built using classical approaches to Conversational AI. Rather than LLMs+RAG, this agent has an special NLU module (natural language understanding) that explicitly predicts and tracks specific named entities in the conversation, allowing for unprecedented control and reliability.