How to build an IC Memo Agent

Workflow Overview

1. User Input

• Node: Message (in-0)

• Purpose: This is where the user enters their question or message.

• Example: “What is the key rationale behind our investment in LinkedIn?”

2. Knowledge Retrieval

• Node: Search SharePoint (OAuth2) (knowledgebase-0)

• Purpose: This node searches a specific folder (“IC Memos”) in your SharePoint for relevant documents or information that can help answer the user’s question.

• How it works:

  • It uses the user’s input as a search query.

  • It retrieves the most relevant document chunks (e.g., from investment memos) and passes them downstream.

3. First AI Analysis

• Node: Anthropic (llm-1)

• Purpose: This node uses an advanced language model (Anthropic’s Claude Sonnet 4) to analyze both the user’s question and the retrieved knowledge base content.

• How it works:

  • It receives the user’s question and the relevant SharePoint content.

  • It generates a thoughtful, context-aware answer, referencing the source documents as needed.

  • The system prompt is tailored for investment committee analysis, ensuring the response is professional and clarifies if information is missing.

4. Output to User

• Node: Response (out-0)

• Purpose: This node displays the answer generated by the Anthropic model directly to the user in the interface.

5. Automated Email Drafting

• Node: OpenAI Agent 2 (llm-2)

• Purpose: This node takes the original question, the Anthropic-generated answer, and the references, and drafts a succinct, professional email summarizing the Q&A.

• How it works:

  • It uses a powerful OpenAI model to compose the email.

  • The email is designed to be sent to the end user, summarizing the question, answer, and references.

  • The node is configured to use a Gmail connection for sending the email (though the actual sending is handled by the tool, not shown in this flow).

How the Nodes Connect

• User input flows to both the SharePoint search and is referenced in the LLM prompts.

• SharePoint search results are fed into the Anthropic LLM for context.

• Anthropic LLM’s answer is:

  • Shown to the user via the Response node.

  • Passed to the OpenAI Agent 2 node for email drafting.

• OpenAI Agent 2 uses the original question, the Anthropic answer, and the references to create a summary email.

Key Points

• Automated, context-rich answers: The workflow combines user input with relevant internal documents for high-quality, reference-backed responses.

• Professional communication: It not only answers the user but also drafts a ready-to-send email summarizing the exchange.

• Seamless integration: The flow leverages both SharePoint for document search and advanced AI models for analysis and communication.