Power Platform Community Forum Thread Details

Hi Anilal,

Yes — Copilot Studio + Power Automate is a viable combination for this POC, but the approach you pick depends on whether you want conversational Q&A over the PDF or a deterministic risk-detection scoring pipeline. Let me break down the supported building blocks from official docs:

1. Generative Answers over uploaded PDFs

You can upload PDFs as a knowledge source at the agent level, and the agent uses generative AI to answer questions over them. This is good for disclosure analysis style queries like "Does this contract contain a confidentiality clause?" or "Summarize the indemnity obligations". Uploaded documents are stored securely in Dataverse, and the agent requires Dataverse search to be enabled in the environment.

Reference: Upload files as a knowledge source – Microsoft Learn

2. User-uploaded files at runtime (interactive POCs)

If users will drop a PDF into the chat for one-off analysis, use a Question node typed as File with Include file metadata enabled, then pass the file into a Power Automate flow or tool. The Power Fx pattern is:

{ contentBytes: Topic.userFile.Content, name: Topic.userFile.Name }

Reference: Pass files to agent flows, connectors, and tools – Microsoft Learn

3. Generative orchestration for legal risk detection

For risk detection specifically, the generative orchestration mode lets the agent reason over knowledge sources, tools, and topics dynamically — useful when you want it to detect specific clause types or flag risky language.

Reference: Orchestrate agent behavior with generative AI – Microsoft Learn

Architectural recommendation for your POC

For legal risk detection specifically, I'd suggest a hybrid:

Use Copilot Studio as the conversational front-end (Question node → File upload).
Pass the file to a Power Automate flow.
In the flow, use AI Builder prompts (or a custom GPT prompt action) with a structured prompt template that returns a JSON schema with detected risks/scores. This gives you deterministic, schema-bound output rather than free-form generative answers — much better for downstream compliance reporting.
Return the structured result back to the agent for the user.

This decouples the conversational UX from the analysis logic and is far more reliable for legal/compliance scenarios than relying purely on the OOTB Generative Answers node.

Hope this gives you a solid POC starting point!

Raghav Mishra — LinkedIn | PowerAI Labs

Found this helpful? Please mark ✅ "Does this answer your question?" so others searching for the same issue can find it quickly. A 👍 on "Was this reply helpful?" or a ♥ Like is also much appreciated!