Setting Up an AI Workflow for Document Prioritization with Detailed Source Citations

Below is one approach to configuring an n8n workflow that meets your specifications. The idea is to build a multi-part workflow where a Chat Trigger starts an AI conversation, an Agent node is used to analyze and prioritize the core five documents, and a custom sub‑workflow or HTTP Request node retrieves additional data (such as meeting transcripts) from Pinecone. All responses are then compiled with clear citations.

---

Workflow Overview

1. Chat Trigger:
   Use the Chat Trigger node as the entry point so that user queries (for example, “What did the team discuss about waiver submissions?”) initiate the process.

2. Agent Node (AI Analysis):
   Configure the Agent node to incorporate a custom prompt that:

   • Specifies the five primary documents (for example, the bill text, associated rules, and supplementary materials).
   • Instructs the AI to prioritize these when analyzing queries.
   • Asks for citations to source documents in all responses.

   Tip: You can use textual context (or even attach the documents’ metadata/identifiers) directly into the prompt so that the agent is “aware” of which documents take priority.

3. Custom n8n Workflow Tool / HTTP Request Node (Pinecone Query):
   For fetching meeting transcripts or additional supporting citations, integrate Pinecone into your workflow. There are two common approaches:

   • Custom Workflow Tool:
     Create a sub‑workflow that the Agent can call. This workflow can use a Custom n8n Workflow Tool node that takes parameters (such as keywords and context) and calls an API for querying Pinecone. In your sub‑workflow, add:

     • An HTTP Request node configured to call Pinecone’s API with the key query (for meeting notes).
     • A function or set of nodes to parse the response so you extract details like file name, speaker, and key discussion points.

   • Direct HTTP Request:
     If you prefer a simpler setup, use an HTTP Request node directly after the Agent node. Pass the extracted query string as a parameter and query your Pinecone collection configured with the documents. Process the response to build citation-backed results.

   Either method works, but using a sub‑workflow can help keep the main workflow clean and modular.

4. Response Formatting and Aggregation:
   Combine the prioritized analysis from the Agent (focusing on the five core documents) with the additional supporting data fetched from Pinecone.

   • Use nodes like “Set” or “Function” to consolidate information.
   • Ensure that the final response includes clear citation metadata (for example, document names or source IDs) for traceability.

5. Output Response:
   Return the comprehensive AI answer to the user. It should include:

   • Primary analysis based on the core documents.
   • Supplementary insights with citations from other documents and meeting transcripts.
   • Clear formatting indicating which parts of the response came from which source.

---

Example Node Configuration Snippet

Below is a simplified pseudo‑configuration for some of these nodes:

YAMLCopied
nodes:
  - name: Chat Trigger
    type: n8n-nodes-langchain.chattrigger
    parameters:
      /* Configure your chat interface options here */
  - name: Agent
    type: n8n-nodes-langchain.agent
    parameters:
      prompt: |
        You have access to five primary documents: [Bill Text, Rules, Supplementary Material A, Supplementary Material B, Supplementary Material C].  
        When answering queries, first analyze these documents. Then, if needed, back up your analysis with citations and additional context from meeting transcripts and notes.  
        Ensure your final response includes clear citations (document names or IDs).
      /* Other parameters to guide AI behavior */
  - name: Fetch Meeting Notes
    type: Custom Workflow Tool / HTTP Request
    parameters:
      url: 'https://YOUR_PINECONE_INSTANCE/query'
      method: POST
      jsonParameters: true
      body:
        query: "waiver submissions"  # This can be dynamic based on user query
        filter: { type: "meetingTranscript" }
      headers:
        Content-Type: application/json
  - name: Response Formatter
    type: Function
    parameters:
      functionCode: |
        // Combine the Agent output with the meeting notes data
        const agentResult = items[0].json;
        const meetingNotes = items[1].json;
        
        // Example: merge the two responses with clear citations
        return [{
          json: {
            answer: agentResult.answer + "\n\nAdditional Meeting Notes:\n" + JSON.stringify(meetingNotes, null, 2),
          }
        }];

Note: Adjust the API URL, parameters, and response parsing to suit your Pinecone configuration.
Also, for successful integration:

• Ensure your Pinecone credentials are properly set up in n8n.
• With n8n cloud, the EXECUTIONS_PROCESS setting may depend on your specific requirements but generally should work with defaults unless you’re running very heavy workflows.

---

Implementation Considerations

• Document Prioritization:
  Embed the identifiers or context for the five key documents in the Agent node’s prompt. This ensures that when the user query comes in, the AI knows to first review these documents.

• Context-Aware Responses:
  The Agent node should be scripted to reference the primary documents first. A follow‑up call (via the Custom Workflow Tool or directly via HTTP) can then fetch additional context for citations if the query requires it.

• Meeting Notes & Pinecone Query:
  Ensure that your Pinecone vector database is pre‑populated with metadata that includes file names, participants, and key discussion points. Use metadata filtering (as suggested in the Pinecone documentation examples) to return structured, citation‑backed information.

• Comprehensive Responses & Citations:
  Use a final formatting step (via a Function node) to amalgamate the primary AI analysis and the additional fetched data. Structure the result clearly so that each citation is linked to its source document.

---

This configuration leverages n8n’s advanced AI components and the modular approach from our examples to deliver a solution that processes and prioritizes documents with citation‑backed output. You can further iterate on prompts and node configurations based on testing and the specific behavior of your AI agent.

Happy workflow building!