graph TD %%{init: {'theme': 'mc','layout': 'elk'}}%% Prompt-vq8my[<div><img src="/_astro/square-terminal.BMOXc-nZ.svg" style="height: 20px !important;width: 20px !important"/></div>Prompt para generar justificación] style Prompt-vq8my stroke:#a170ff OpenAIModel-d3dpa[<div><img src="/_astro/openAI.BhmuxEs3.svg" style="height: 20px !important;width: 20px !important"/></div>Genera Justificación] style OpenAIModel-d3dpa stroke:#a170ff Prompt-tma9i[<div><img src="/_astro/square-terminal.BMOXc-nZ.svg" style="height: 20px !important;width: 20px !important"/></div>Prompt] style Prompt-tma9i stroke:#a170ff OpenAIModel-1tbks[<div><img src="/_astro/openAI.BhmuxEs3.svg" style="height: 20px !important;width: 20px !important"/></div>Genera Respuesta] style OpenAIModel-1tbks stroke:#a170ff ChatInput-6u0w8[<div><img src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Input] style ChatInput-6u0w8 stroke:#a170ff CombineText-hw2cy[Combina la justificacion con la respuesta] style CombineText-hw2cy stroke:#a170ff ChatOutput-k57y4[<div><img src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output] style ChatOutput-k57y4 stroke:#a170ff Prompt-vq8my -.- OpenAIModel-d3dpa linkStyle 0 stroke:#a170ff OpenAIModel-d3dpa -.- Prompt-tma9i linkStyle 1 stroke:#a170ff Prompt-tma9i -.- OpenAIModel-1tbks linkStyle 2 stroke:#a170ff ChatInput-6u0w8 -.- Prompt-vq8my linkStyle 3 stroke:#a170ff ChatInput-6u0w8 -.- Prompt-tma9i linkStyle 4 stroke:#a170ff OpenAIModel-1tbks -.- CombineText-hw2cy linkStyle 5 stroke:#a170ff CombineText-hw2cy -.- ChatOutput-k57y4 linkStyle 6 stroke:#a170ff OpenAIModel-d3dpa -.- CombineText-hw2cy linkStyle 7 stroke:#a170ff

🧩 Overview

This workflow automatically produces a fully‑justified answer to a user’s question.
It first generates a rationale, then integrates that rationale into a final response before delivering the output back to the chat interface.

⚙️ Main Features

Generates a logical justification for each user query using an LLM.
Combines the justification with the final answer into a single coherent message.
Uses two separate OpenAI models: one for rationale creation and one for answer generation.
Supports dynamic prompt templates that inject user questions and generated rationales.
Outputs the final answer directly to the chat session for immediate display.

🔄 Workflow Steps

Component Name	Role in the Workflow	Key Inputs	Key Outputs
Chat Input	Receives the user’s question and session metadata.	User message, conversation ID, session ID	Structured chat message
Prompt – Rationale	Constructs a prompt asking the model to produce a rationale.	Question from chat message	Prompt text
OpenAI Model – Rationale	Generates a rationale text from the prompt.	Prompt text	Rationale text
Prompt – Final Answer	Builds a prompt that includes the question and the generated rationale to ask for the final answer.	Question, Rationale text	Prompt text
OpenAI Model – Final Answer	Produces the final answer text.	Prompt text	Answer text
Combine Text	Concatenates the rationale and answer using a newline delimiter.	Rationale text, Answer text	Combined justification‑answer text
Chat Output	Sends the combined text back to the chat session as the bot’s reply.	Combined text, conversation ID, session ID	Chat message displayed to the user

🧠 Notes

The first OpenAI model is tuned for producing concise rationales; the second model focuses on crafting the final answer.
Both models use the same “gpt‑4o‑mini” engine, but the architecture can be swapped out by changing the model name field.
Prompt templates are dynamic; placeholders ({question}, {rationale}) are replaced automatically at runtime.
The delimiter for combining text is two newline characters (\n\n) by default, ensuring clear separation between rationale and answer.
Session and conversation identifiers maintain state across multiple turns, allowing context‑aware replies.
Error handling for model failures is not explicitly modeled; in production, the workflow should capture and surface failures to the user or a monitoring system.