How Do I do this: "Prompt Tree" Decision-based Interaction Module

[howtodoml]

Objective:
Introduce a structured interaction mechanism, named "Prompt Tree", to enhance our chatbot's user experience by offering decision-driven engagements.
Background:
In our pursuit to streamline user interactions and provide a guided experience, we've identified the need for a decision tree-based approach. This methodology aligns with our objective to offer users clear, concise, and context-driven conversations while efficiently utilizing the capabilities of LLM.
Requirements:

1. Dynamic Branching:
   • Create a logic-driven pathway system, where the chatbot will present options or responses based on previous interactions.
   • Branches should be easily configurable to cater to different user journeys.
2. Data Collection & Retention:
   • At distinct points in the interaction, the system must have the capability to capture, retain, and reference user input.
   • This captured data should be available for reference in any subsequent node, and its lifecycle should be defined based on the session's duration or relevance.
3. Built-in Validators:
   • Introduce checkpoints at specified nodes to ensure that the chatbot's outputs are aligned with expected outcomes.
   • This will act as a quality control mechanism, ensuring the consistency of user interactions.
4. Scratchpad Synergy:
   • While the primary goal is to reduce dependency on the scratchpad, its integration should remain seamless. This will ensure that when necessary, the LLM can reference the scratchpad effectively.
5. Visual Designer Interface:
   • Empower our chatbot developers with a graphical tool to visualize, design, and modify the Prompt Tree.
   • This interface should also allow for debugging and performance analytics.

Use Case Illustration:
Consider our customer support bot:

1. Initial Prompt: "How may I help you today?"
2. User Response: "Billing"
   • Chatbot Branches to:
     • "Would you like to pay your bill?"
     • "Do you want to check your bill status?"
     • "Is there an issue with your bill?"
       Each subsequent branch narrows down the user's intent, offering a more tailored response or solution.

[alexchaomander]

@hathind-ms @alliscode @evchaki This is a customer ask that could be interesting to see expressed through Chat Copilot.

[markwallace-microsoft]

I am cleaning up old unanswered discussions