Best Practices

Based on our experience at Prst.ai, we've identified several best practices to help you succeed in prompt engineering:

Evaluate Prompts with SMEs

Involving subject matter experts (SMEs) significantly reduces the risk of generating invalid or irrelevant outputs, thereby enhancing prompt quality and the business value of your product. While SMEs may not need to know the details of the prompt itself, presenting only the generation result during experiments can keep the process clear and focused.

Use Version Control

Usage of prompt version control enables you to store changes history and information about evaluations at each stage of the prompt creation lifecycle. Creating a new prompt version during each Prompt Loop iteration allows for easy comparison and understanding of which version performed better.

💡 Note: Read More about Version Control in Prompts Section

Regular Re-Evaluation:

Even if you believe you have the best prompt that meets all your requirements, it's still important to regularly re-evaluate it. Testing prompts with a wide variety of input parameters is crucial to uncovering any potential issues or incorrect results. Without comprehensive testing, unpredictable outcomes may occur.