n8n Workflow: AI Vision Chatbot for Deal Flow with Qdrant Vector Store

This n8n workflow creates an intelligent AI chatbot that can process pitch decks (images), extract information, and answer questions based on the content, leveraging a Qdrant vector store for efficient knowledge retrieval. It's designed to streamline deal flow analysis by providing an interactive way to query and understand pitch deck data.

What it does

This workflow orchestrates a sophisticated AI-powered interaction:

1. Listens for Chat Messages: It triggers when a new chat message is received, initiating a conversation.
2. Initial Information Extraction (Optional): It attempts to extract structured information from the chat message using an AI-powered information extractor, potentially to understand the user's intent or key entities.
3. Processes Pitch Deck Image (Conditional): If an image (presumably a pitch deck slide) is provided in the chat, it performs the following:
   • Loads Image Data: Converts the image binary data into a format suitable for processing.
   • Performs Image-to-Text Conversion: Uses an AI Agent with vision capabilities to describe the image content.
   • Splits Text into Chunks: Divides the extracted text into smaller, manageable chunks for embedding.
   • Generates Embeddings: Creates vector embeddings for each text chunk using OpenAI's embedding model.
   • Stores in Qdrant: Upserts these embeddings into a Qdrant vector store, making the pitch deck content searchable.
4. Answers Questions with Vector Store: Utilizes a Vector Store Question Answer Tool to retrieve relevant information from the Qdrant store based on the user's query.
5. Maintains Conversation Context: Incorporates a simple memory buffer to keep track of the conversation history, allowing for more natural and coherent interactions.
6. Responds to User: Constructs a response using an OpenAI Chat Model, incorporating information from the vector store and conversation history.
7. Handles No Image Scenario: If no image is provided, it can still interact using the AI Agent and memory, potentially answering general questions or guiding the user to upload a pitch deck.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running n8n instance (self-hosted or cloud).
• Airtable Account: Used for the "Airtable Trigger" and "Airtable" nodes, suggesting potential integration with a deal tracking database.
• OpenAI API Key: For the Embeddings OpenAI and OpenAI Chat Model nodes, enabling text embeddings and AI chat capabilities.
• Qdrant Instance: A running Qdrant vector database instance for storing and retrieving document embeddings.
• LangChain Integration: The n8n LangChain nodes package (@n8n/n8n-nodes-langchain) must be installed and configured in your n8n instance.
• Image Processing Capabilities: The Edit Image node is present, implying image manipulation might be part of the flow (though its current connections are not fully defined in the provided JSON, it's a capability).
• Code Node Execution: The Code node implies custom JavaScript execution, which might require specific environment configurations or dependencies.

Setup/Usage

1. Import the Workflow: Download the provided JSON and import it into your n8n instance.
2. Configure Credentials:
   • Set up your Airtable credentials.
   • Configure your OpenAI API Key credentials.
   • Set up your Qdrant credentials, including the host, port, and API key if required.
3. Configure Trigger Nodes:
   • Airtable Trigger: Configure the Airtable base, table, and trigger event (e.g., new record) if you intend to use Airtable as a primary data source for deal flow.
   • Chat Trigger: Ensure your chat platform (e.g., Telegram, Slack, Discord) is integrated with n8n and configured to send messages to this workflow's Chat Trigger webhook.
4. Customize AI Agent and Chains:
   • AI Agent: Review and customize the AI Agent node, especially its tools and prompts, to define how it should interpret images and interact.
   • Information Extractor: Adjust the schema and prompt for the Information Extractor node to define what structured data you want to extract from initial messages.
   • OpenAI Chat Model: Configure the model (e.g., gpt-4, gpt-3.5-turbo) and any specific parameters for the OpenAI Chat Model.
5. Deploy and Activate: Once all credentials and configurations are set, activate the workflow.

This workflow provides a powerful foundation for building an interactive AI assistant to manage and analyze pitch deck information, significantly enhancing deal flow processes.