🚧

UnifyAI is in Alpha

In Alpha, you can expect that UnifyAI is growing and iterating rapidly (even multiple times per day). Some of this documentation may be lagging a bit behind the rapid iterations. We are working hard to keep up with the data science team! If you are having any issues, you can reach us any time at [email protected]

Importance of prompt engineering for AI products.

Well-crafted prompts guide the model's output, allowing developers to fine-tune performance, address biases, and align with user expectations.

You can think of prompt engineering like giving directions to a friend – it's the secret sauce for a smooth conversation with AI. The prompts you use shape how the AI responds, making it important to choose them wisely. By tweaking these prompts, developers can fine-tune the AI's behavior, fix biases, increase accuracy, and make sure it's on the same page with you.

How to create effective prompts

Crafting effective prompts is often an iterative process to get the right level of clarity, specificity, and experience. It is recommended to keep these items in mind:

1. Define your task: Clearly picture the specific task or field that you want UnifyAI to generate. For example, if you are running UnifyAI on forum content, you may want a metadata field of a post author's engagement metrics.
2. Specify an output: Which field would you like UnifyAI to generate? For example: "post.engagement"
3. Use clear language: The clearer the language, the better the results. In the examples of author's posting engagement; there may be multiple authors or a difference between the author of a post versus the author of a thread.
4. Iterate: If you are not getting a clear answer, try adding more specific information.
5. Increase testing: Try to fool UnifyAI, and present multiple variances to ensure the response matches your goal.

📘

Prompt examples

Examples of prompts are a great place to start. Review "Getting started with UnifyAI" for examples of prompts.

Getting support in prompt creation

Your Datastreamer team is always ready to support prompt creation. Just reach out to us! If you are not sure who your Datastreamer team is, reach out to [email protected] and we can check for you.