Power Platform Community Forum Thread Details

I’m building a multi-agent solution in Copilot Studio where users upload multiple documents when they are conversing with the parent agent(PDF, Word, Excel,ppt etc), and the system extracts content, normalizes it into structured JSON, and passes it to downstream agents for risk analysis.

The problem is I don't know how to enable the Parent agent to ingest all different types of file and then extract the data and pass it on to child agents. I am a total newbie to creating agents on Copilot Studio and would really appreciate if someone can solve this. I have tried creating a power automate flow to manage this but couldn't figure out. For context, multiple users would upload multiple types of files when this is in production and I only have access to standard connectors.

Categories:

Building Copilot Studio chatbots in Microsoft Teams

Hi @BP-21041620-0,

Based on the given context (you only have access to standard connectors), I understand why you have tried with PowerAutomate instead of using Azure Form Recognizer or OCR services for content extraction. However - it is possible where you have left of.

Given that you have only standard connectors available and need to enable your parent agent to ingest multiple file types (PDF, Word, Excel, PowerPoint), extract their content, and pass structured data to child agents, here’s a cpractical, production-ready approach you can implement within Copilot Studio and Power Platform:

To me this is the flow for your use caes, you need to achieve:

File Upload in Parent Agent: Configure the parent agent’s conversation to accept file uploads. When users upload files, save them to a SharePoint document library or Azure Blob Storage using standard connectors.

Trigger Extraction via Power Automate: Create a Power Automate flow triggered by the file upload event in SharePoint or Blob Storage. The flow will process each uploaded file based on its type.

Content Extraction Using Standard Connectors:

For Word, Excel, PowerPoint: Use the Microsoft Graph API connector or Office 365 connectors in Power Automate to extract text and metadata.
For PDFs: Use the AI Builder Form Processing model (available as a standard connector) or the OneDrive for Business connector to extract text if OCR is needed.
If AI Builder is not available, consider extracting text via Power Automate’s built-in PDF text extraction or third-party connectors if allowed.

Normalize Extracted Data: In the flow, transform the extracted content into a consistent JSON schema.This can be done using Power Automate’s data operations (Compose, Parse JSON, etc.).

Pass Data Back to Parent Agent or Directly to Child Agents:The normalized JSON can be sent back to the parent agent via a custom API call or stored in Dataverse or SharePoint for the parent agent to retrieve.Alternatively, the flow can invoke child agents directly by calling their endpoints or triggering their topics with the JSON payload.

Orchestration and User Feedback: The parent agent can notify users about extraction progress and results. Handle errors gracefully and allow users to re-upload or correct files if needed.

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!