Assistant only carries the first message inside the dashboard. Questions are manually fed through system calls, and I am currently implementing a sandwich pattern where the first part includes the personality and temperature of the assistant, the middle part contains the questions, the last part contains the end call message transition. For each question also implements a sandwich pattern that adjusts the requirements for that question and if there are any other follow ups. Insights provided for the sandwich are all generated by AI converted from an initial user prompt. Questions are being asked correctly and relevant data are being collected but the underlying issue is sometimes that the caller is not finished speaking yet and the assistant jumps to the next question. What I tried so far - using the generated context of the AI as a System Prompt in VAPI (Skipping client fed questions) - Removing the narrative of insights from the sandwich pattern as well as the questions, and only ask the assistant to ask the defined questions - Adjusting Speaking plan via dashboard ! No satisfying results so far, same behavior What I am trying to experiment on - Tool Functions