How to build a Proposal Reference Agent

Workflow Overview

1. User Inputs

• Proposal Topic Input (in-0)

  • Purpose: The user enters the topic for a new proposal (e.g., “Identify proposals for a highway expansion project”).

• Recipient Email Input (in-1)

  • Purpose: The user provides the email address to which the proposal summary and references will be sent.

2. Knowledge Base Search

• Google Drive Knowledge Base (knowledgebase-0)

  • Purpose: Searches a designated Google Drive folder (“Proposals”) for documents relevant to the entered proposal topic.

  • How it works: Uses the topic from in-0 to semantically search and retrieve the most relevant proposal documents (e.g., PDFs about highway expansion, bridge rehabilitation, etc.).

3. Proposal Summarization

• OpenAI LLM (llm-0)

  • Purpose: Summarizes the key points of the most relevant proposal(s) found in the knowledge base.

  • How it works: Takes the topic from in-0 and the content from knowledgebase-0, then generates a concise summary.

  • Output: A human-readable summary and a list of citations (references to the original proposal documents).

4. Full Proposal Extraction

• Perplexity LLM (llm-1)

  • Purpose: Extracts the full plain text of the referenced proposals for inclusion in the email.

  • How it works: Receives the summary and citations from llm-0, then uses a tool to fetch the actual proposal documents from Google Drive and converts them to plain text.

5. Email Sending

• Send Proposal Reference Email (action-2)

  • Purpose: Sends an email to the recipient with the proposal summary and the full referenced proposals attached or included.

  • How it works:

    • Recipient: Uses the email from in-1.

    • Subject: References the proposal topic from in-0.

    • Content: Includes the summary from llm-0 and the full text from llm-1.

    • Attachments: Includes the referenced proposal documents.

    • Provider: Uses Gmail to send the email.

6. Output

• Output Node (out-0)

  • Purpose: Indicates that the workflow has completed and the email has been sent.

Summary Table

How Data Flows

1. User provides a topic and recipient email.

2. The system searches Google Drive for relevant proposals.

3. A summary of the best-matching proposal(s) is generated.

4. The full text of the referenced proposals is extracted.

5. An email is sent to the recipient with the summary and full proposals.

6. The workflow signals completion.