Best Practices

1. Define and Register Prompts Clearly:

   1. Group similar prompts into families.

   2. Use clear naming conventions for prompt keys.

2. Log Consistently:

   1. Ensure every LLM call is logged with associated prompts, input data, and output responses.

   2. Regularly review logs via the Data Viewer to maintain high visibility into system performance.

3. Monitor Performance Regularly:

   1. Use Performance Over Time to track trends and quickly identify issues.

   2. Set up alerts for when prompt performance dips below your target threshold (e.g., 98%).

4. Utilize Adaptive Re-Prompting:

   1. Configure threshold settings based on your application's performance needs.

   2. Test and iterate to find the best-performing prompt variants.

5. Leverage Visual Analytics:

   1. Use the Show Input Space and Prompt Performance tools to gain insights into your prompt efficacy.

   2. Make data-driven decisions to continuously refine your prompts.

6. Integrate User Feedback:

   1. Combine in-product feedback (e.g., thumbs up/down) with system analytics to validate prompt performance.

   2. Regularly update your prompt families based on both quantitative scores and qualitative feedback.

7. Ensure Data Privacy and Ethical Standards:

   1. Follow best practices for data handling to ensure sensitive information is secured.

   2. Consider on-premise or containerized deployment if necessary for privacy compliance.