AI Copilot System Design Workflow Documentation (Schema to Draft Editor)

[Grinzypino]

The draw.io file is present in the main directory for a better view and control over the design made.

[Grinzypino]

AI Copilot System Design Workflow Documentation

Overview

This documentation explains the workflow of the system design as depicted in the provided diagram. The process begins by sending a JSON schema path to the /generate_q API endpoint and concludes when the workflow reaches the Draft Editor.

Workflow Steps

1. Starting Point: JSON Schema Path to /generate_q

   • The workflow starts by sending a JSON schema path to the /generate_q endpoint.
   • API Endpoint: /generate_q
   • Action: This endpoint generates a list of questions based on the provided JSON schema.

2. Question Generation (/generate_q)

   • This endpoint calls certain functions that call GPT 3.5 Turbo to generate questions for the JSON schema and also creates a current JSON which is initialised with null or empty list values.
   • Output: A set of questions in JSON format, current_json object and a dictionary of expected answers in JSON format.

3. Question Display and User Interaction

   • The generated questions are converted into Hindi and displayed on a Q&A form for the user to fill out.
   • UI Component: Converts questions to Hindi and presents them to the user.
   • User Action: The user provides answers to the displayed questions in Hindi.

4. Answer Checking (/check_ans)

   • The user's answers are sent to the /check_ans endpoint.
   • API Endpoint: /check_ans
   • Action: This endpoint checks the user’s answers for correctness.

5. Reframing Questions (/reframe_q)

   • If the user’s answers are not satisfactory or if additional clarification is needed it returns a dictionary of only those questions and the /reframe_q endpoint is invoked by UI.
   • API Endpoint: /reframe_q
   • Action: This endpoint reframes the questions to elicit better or more detailed responses from the user.
   • The user is then presented with the reframed questions, and the process returns to the answer-checking step.

6. Draft Generation (/generate_draft)

   • If all questions have been satisfactorily answered then UI recieves an empty dictionary of questions that needed to be reframed and the /generate_draft endpoint is invoked to create a draft document.
   • API Endpoint: /generate_draft
   • Action: This endpoint generates a draft based on the user’s answers.

7. Draft Editor

   • The generated draft is sent to the Draft Editor for final review and editing.
   • Component: Draft Editor
   • Action: The user can review, edit, and finalize the draft document.

This workflow ensures that the questions derived from a JSON schema are thoroughly answered, checked for correctness, and used to generate a comprehensive draft document, which can then be finalized in the Draft Editor.