LLM Integration for Guide Creation Assistance

[wavebreaker-lucas]

Objective:
Integrate a Large Language Model (LLM) to assist in generating step titles, descriptions, and summaries for step-by-step guides. Screenshots are automatically captured via the platform’s workflow recording tool, and the LLM will assist in generating content based on these screenshots. Additionally, include default rule-based generation for titles and descriptions if AI assistance is not used.

Subtasks:

• LLM Model Selection:
  Choose and integrate a suitable LLM (e.g., OpenAI GPT-4, Anthropic Claude, Azure OpenAI API) to assist with content generation.

• Default Rule-Based Title & Description Generation:
  Implement default rules for generating step titles and descriptions when AI is not utilized.
  Default title rules: Generate titles based on the workflow.
  Default description rules: Generate basic descriptions using a standard template, such as "Click on X to perform Y."

• AI-Assisted Step Description Generation (via Automatic Screenshot):
  Use the automatically captured screenshots along with metadata (e.g., the action performed, UI elements) to generate step descriptions with the LLM.
  The LLM generates descriptive, context-aware step instructions based on the content of the screenshot and user actions.
  Users can choose to accept, refine, or edit the AI-generated descriptions.

• AI-Assisted Title Generation (based on Step Description):
  Use the LLM to generate step titles based on the AI-generated step descriptions.
  Users can select either the AI-suggested title or rely on the default rule-based title if they prefer.

• Summary Generation (based on Titles & Descriptions):
  The LLM generates a guide summary by analysing the titles and step descriptions.
  Users can edit or regenerate the summary based on their preferences.

• UI/UX Integration:
  Integrate user-friendly buttons for generating AI-assist titles, descriptions, and summaries.
  Provide a clear option for users to fall back on default rule-based generation if they choose not to use AI.
  Display AI-generated suggestions in a sidebar or pop-up for easy review and insertion.

• Backend Integration:
  Connect with the LLM via API for scalable AI-assisted content generation.
  Implement default rule-based generation on the backend as a fallback option.
  Ensure rate limiting and caching for handling high-traffic AI requests efficiently.

• Security & Privacy:
  Ensure data security and compliance with privacy regulations (e.g., GDPR, CCPA) for any data processed, including screenshots and AI-generated content.

Acceptance Criteria:

• LLM generates step descriptions based on screenshots and titles based on descriptions.
• Users can choose between AI-generated or rule-based titles/descriptions and can regenerate or edit suggestions.
• The system is scalable, handles high traffic, and complies with privacy regulations.

Labels: LLM-integration, content-generation, titles, summaries, AI-assistance, rule-based
Milestone: LLM Assistance