As a dev team, we want to ensure robust error handling and JSON validation for our backend services.

[Endlessflow]

As a dev team, we want to ensure robust error handling and JSON validation for our backend services.

Context:

Currently, our backend services rely on two external services during the analysis route: Document Intelligence and Azure OpenAI (GPT-4). Failures in these services can lead to unhandled errors, invalid data, and potential system instability. We need to ensure that our backend can gracefully handle service unavailability, failures, and invalid data responses. Additionally, it's crucial to validate the JSON output from our processing pipeline to ensure it meets the required structure and fields before sending it to the frontend.

Child Issues:

• [x] https://github.com/ai-cfia/fertiscan-backend/issues/69
• [x] https://github.com/ai-cfia/fertiscan-backend/issues/43
• [x] https://github.com/ai-cfia/fertiscan-backend/issues/72

Acceptance criteria:

• Error Handling for Document Intelligence Service:

  • The system should detect if the Document Intelligence service is unavailable and handle the error gracefully without crashing.
  • Implement fallback mechanisms or retry logic with a limited number of retries.
  • Log the error details for further analysis.

• Error Handling for LLM Service:

  • The system should detect if the LLM service fails or returns invalid data (non-JSON format) and handle the error gracefully.
  • Implement fallback mechanisms or retry logic with a limited number of retries.
  • Log the error details for further analysis.

• JSON Output Validation:

  • Validate the JSON output from GPT to ensure it contains the required fields and follows the correct structure.
  • If the JSON is invalid, handle the error gracefully and provide meaningful error messages.
  • Implement unit tests to verify the JSON validation logic.

[snakedye]

It seems like Pydantic as is already handles the edge cases of a JSON extra keys or missing keys.