M7.01 - Implementing context in agents using storage buckets
What will this agent do?
Organizations frequently receive employee support requests that span multiple areas—from IT troubleshooting to HR related questions. In this exercise, you’ll create an employee support agent that analyzes the employee’s query, determines the appropriate issue category, maps it to the correct support team using a CSV file stored in a storage bucket, and then provides a suitable response to help move the issue toward resolution.
All required inputs—such as the system prompt, user prompt, support group file, and test input file—are included in the downloadable zip package
Employee support agent.zip:
https://academy.uipath.com/content/ac...
Key takeaways:
Agent structure and configuration
System prompt:
Defines the agent’s purpose, the tasks it must carry out (such as identifying employee issues and mapping them to the correct support group using contextual data), and the order in which those tasks should be executed.
Input arguments:
Values provided at the start to guide the agent’s behaviour. In this scenario, the inputs are the user’s Name and their Message.
Output arguments:
The information the agent returns after completing its process. Here, that may be the appropriate support group (when identifiable) or a request for additional details.
User prompt:
Shapes how the agent interacts with employees—managing greetings, asking clarifying questions, and responding based on the support data available.
Context grounding with storage buckets
• The support group information comes from a CSV file that maps topics and services to their corresponding support teams.
• This file is stored in a UiPath Orchestrator storage bucket, providing a centralized and reusable source of contextual knowledge.
• To make the CSV usable by the agent, an index is created with the storage bucket as its data source.
• Adding this index to the agent’s context allows it to generate responses grounded in internal organizational information.
Test and validate
The agent is evaluated using real user queries, including both entries that match the support group list and those that do not. Testing verifies that the agent correctly identifies the appropriate support group when possible and provides a suitable fallback response when it cannot.
Context grounding in UiPath.pdf:
https://academy.uipath.com/content/ac...