Prompt Pattern 2 - Analyst / Doer

Refine your prompt and alter LLM outcomes with Generative AI parameters for personalized text in the Performer process. This accelerator generates data from summary emails.

Prompt engineering is an artificial intelligence (AI) concept that optimizes the prompts and parameters of a large language model (LLM) to produce precise, context-aware responses. The Prompt Pattern 2 – Analyst / Doer Solution Accelerator uses this approach to fine-tune the prompts and retrieve accurate responses from a LLM. Each iteration’s output is clearly presented, enabling the assessment of the AI’s accuracy. Modifications in each iteration can be made to the prompt, AI parameters, AI model, or even AI provider. The engineer can verify each LLM response manually before it’s used in other automations. If satisfied, the engineer can run the prompt against all queue items without additional manual checks.

This Solution Accelerator exemplifies generating structured data from a summary email. Initial context is gathered from a user's email inbox that receives an email summarizing a virtual meeting. A structured output, which specifies what UiPath process to call and any arguments, is created using the context. After refinement and verification, separate paths of processing occur in the Performer. In this instance, workflows in the LLM can include scheduling a follow-up meeting, sending specific documents for review to participants, and forwarding the summary to all participants.

• Unattended and Attended Dispatcher Templates are used to gather unstructured context from multiple sources.
• Automatically insert context into a prompt.
• Human-in-the-loop validation of LLM responses. Users can change prompts, parameters, models, and providers in one verification screen.
• Automatically run against all gathered in a queue or manually verify each LLM response.
• Unattended and Attended Performer Templates are used to execute processes based off the structured LLM response.