How to build a Policy Q&A and Coverage Validation Agent

Workflow Overview

1. User Input

• Node: Text Input

• Purpose: Collects the user’s insurance policy question.

• Example: “Are home improvements covered?”

2. Document Upload (Knowledge Base)

• Node: Document Upload

• Purpose: Allows the user to upload their insurance policy document(s).

• How it works: The uploaded document is indexed and made searchable for relevant information.

3. LLM (AI Answer Generation)

• Node: OpenAI

• Purpose: Answers the user’s question using only the information found in the uploaded policy document.

• How it works:

  • The AI is instructed to:

    • Only use information from the uploaded policy.

    • Always cite the relevant section, clause, or page number.

    • Respond with “Not specified in the policy.” if the answer isn’t clear.

    • Indicate if the policy is ambiguous.

    • Avoid legal advice or personal opinions.

  • The AI receives:

    • The user’s question.

    • The searchable policy text from the knowledge base.

4. Output

• Node: Output

• Purpose: Displays the AI’s answer to the user, including any citations from the policy document.

How the Data Flows

1. User submits a question →

2. Question is sent to the Document Upload node (to help form a search query) →

3. Document Upload node provides relevant policy text →

4. LLM (OpenAI) receives both the user’s question and the relevant policy text →

5. LLM generates a policy-based answer, citing the document →

6. Output node displays the answer to the user

Visual Summary

In short:
The user asks a question, uploads their policy, and the AI answers strictly based on the uploaded document—always citing the source or stating if the answer isn’t specified.