AI App
Last updated
Last updated
Dataworkz RAG Applications can be accessed from within Dataworkz Agents. This provides the Agent with access to a powerful RAG based system to answer questions from unstructured document sources such as PDFs, HTMLs, doc files, etc. Dataworkz RAG Applications provide the ability to take in a text as a query and respond with a relevant answer generated by an LLM using RAG based techniques. E.g. What products cannot be returned? Can be asked of a policy document and the RAG Application will return with the exact set of rules that answer that question (assuming that the answer is present). In addition, Dataworkz RAG Applications also support additional filters based on metadata - a very powerful concept that allows for the data corpus to be partitioned to provide more retrieval accuracy.
An Dataworkz RAG Application must be created in the same account
To setup an AI App tool that uses a Dataworkz RAG Application -
Select AI Agents > Create a Tool > AI App
Name: Provide a good name to the tool
Description: Provide a useful description of the tool - usually this is the description of what this tool can be used for - e.g. use this tool to answer questions about store policies OR this tool can answer troubleshooting questions
AI Application: Select the correct RAG Application from the list
LLM: Pick the LLM you want to use to answer questions
Input Parameters: The default parameter is automatically added and cannot be changed (although its description can be updated to add more information). If you have any metadata setup in your RAG Application, you can select them here as additional input parameters
Output: The output of the tool is pre-determined however you can add additional information in the description
Examples: These are questions that were asked in this RAG Applications - these can help the Agent understand the RAG Application better
Use this tool when you want to provide the Agent with unstructured data. The Agent can use it for a variety of purposes - glossary of terms, to get rules, to understand approaches to solve problems (such as troubleshooting), etc.
