Hyperflow Prompt Optimization
Use the Optimization command on the Prompt Details window for a Hyperflow to increase the efficiency of your hyperflow prompts. Prompt Optimization improves consistency, accuracy, and repeatability across runs, contexts, and users, without requiring any prompt-writing expertise.
To use the Hyperflow Prompt Optimizer:
Navigate to AI Automation > Hyperflows in the Aisera Admin UI.
Make sure the correct application/bot is shown in the upper-left corner of the Hyperflows window.
Choose one of the Hyperflows to open the Hyperflow Details window for the chosen hyperflow.
Open a Hyperflow and click the pencil icon in the upper-right side of the window to make any changes.
Choose the pencil icon next to the word Prompt at the top left of the window. This action brings you to the Hyperflow Prompt Editor.
Click the pull-down menu next to the three horizontal dots above the right side of the prompt editor canvas and choose Optimize.
This action causes the Aisera Gen AI Platform to optimize the content in your prompt editor canvas.
Once optimized, you can:
Review the improved prompt
Edit it further if needed
Revert to the previously saved version at any time
To revert the optimization, open the three-dot menu again and select Revert.
App Instructions Prompt Supports Hyperflows
The App Instructions prompt now supports Hyperflows in the [Action] field, in addition to standard workflows. When a Hyperflow exists that matches a user Request, it will be identified and returned automatically. If both a Hyperflow and a Workflow exist for the same query, the Hyperflow takes precedence, allowing you to take advantage of more powerful, complex workflows where available.
To configure App Instructions, navigate to LLMs > Prompts Studio in the Aisera Admin UI and select the convai_app_instructions_prompt_gpt4_chat_completions prompt.
Select Prompt
Select (double-click) the prompt to open the Prompt Playground that allows you to customize and modify the prompt.
System message field in Prompt Playground
Enter information between the <actions></actions> tags within the code in the System message field.
Last updated
Was this helpful?
