Prompt engineering #377

[scezen]

Issue Reference

This PR addresses issue #377.

Objective

1. Merge GenAI prompts templates into a single "meta template" to streamline multiple text generations with embedded prompts in a single call.

Summary of Changes

• ApiConfig: Updated GenAiCompletionService bean to include PromptConfig as parameter. Added PromptConfig bean.

• GenAiCompletionService: Updated and added applyRateLimit method for rate limiting.

• AiService: Refactored and enhanced following clean code principles:

  • Renamed prompt method to generatePromptResponse for clarity.
  • Updated complete method to use the new generateDescriptionsForProduct method.
  • Added methods:
  • generateDescriptionsForProduct: Handles the generation of descriptions for a product.
  • shouldSkipDescriptionGeneration: Checks if description generation should be skipped based on existing data and configurations.
  • createAiDescriptions: Generates AI descriptions for a product using the root prompt and individual prompts.
  • storeGeneratedDescriptions: Stores the generated AI descriptions in the product.

• Configuration Files: Created AiPromptsConfig to handle root prompts and list of PromptConfig. Added rateLimitDelay attribute to PromptConfig.

• tv.yml: Updated aiConfigs for the new meta template structure.

Additional Notes

• The generatePromptResponse method ensures a consistent style for all generated texts.
• The system checks if a text already exists to avoid unnecessary regenerations unless forceOverride is enabled.
• Rate limit delay can be configured in PromptConfig to manage timing between API calls.

[GoulvenF]

Thanks @scezen for this work ! Document and code are excellent, congrats.