n8n Workflow: Document Analysis and Chatbot Creation with Llama-Parser, Gemini LLM, and Pinecone DB

This n8n workflow provides a comprehensive solution for analyzing documents, extracting information, and building a chatbot using advanced AI components. It leverages Langchain nodes to integrate with various AI services, including a Llama-Parser for document loading, Google Gemini for language modeling, and Pinecone for vector database management.

What it does

This workflow automates the following key steps:

1. Triggers on Form Submission: Initiates the workflow when a user submits data via an n8n form.
2. Loads Document Data: Takes the submitted document (expected as binary data) and loads it using a Default Data Loader.
3. Splits Document into Chunks: Uses a Recursive Character Text Splitter to break down the document into manageable chunks for processing.
4. Generates Embeddings: Creates vector embeddings for each document chunk using the Mistral Cloud Embeddings service.
5. Stores Embeddings in Pinecone: Uploads the generated embeddings to a Pinecone Vector Store for efficient similarity search.
6. Triggers on Chat Message: Waits for incoming chat messages to initiate the chatbot interaction.
7. Retrieves Relevant Information: Uses a Vector Store Retriever to fetch relevant document chunks from Pinecone based on the chat message.
8. Answers Questions with Gemini LLM: Processes the retrieved information and user query through a Google Gemini Chat Model to generate a coherent answer.
9. Extracts Structured Information (Optional): Can extract structured information from the document using an Information Extractor.
10. Sends Email (Optional): Includes a placeholder for sending emails, potentially for notifications or human-in-the-loop approvals.
11. Performs HTTP Requests (Optional): Contains a placeholder for making HTTP requests, which could be used for external API calls.
12. Conditional Logic (Optional): Features an If node for implementing conditional branching based on workflow data.
13. Loops Over Items (Optional): Includes a Loop Over Items node for iterating through collections of data.
14. Outputs Markdown (Optional): Can format output into Markdown.
15. Executes Custom Code (Optional): Allows for custom JavaScript logic execution.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running n8n instance.
• Pinecone Account: An API key and environment for Pinecone.
• Google Gemini API Key: Access to the Google Gemini LLM.
• Mistral Cloud Account: API access for Mistral Cloud Embeddings.
• Gmail Account (Optional): Configured Gmail credentials for sending emails.
• Langchain Nodes: Ensure the @n8n/n8n-nodes-langchain package is installed in your n8n instance.

Setup/Usage

1. Import the Workflow: Download the provided JSON and import it into your n8n instance.
2. Configure Credentials:
   • Set up your Pinecone credentials (API Key, Environment, Index Name).
   • Configure your Google Gemini Chat Model credentials (API Key).
   • Set up your Mistral Cloud Embeddings credentials (API Key).
   • (Optional) Configure your Gmail credentials if you plan to use the email functionality.
3. Configure the Form Trigger:
   • The "On form submission" node acts as the entry point for document uploads. Ensure the form is configured to accept binary data (e.g., a file upload field).
4. Test Document Analysis:
   • Submit a document through the n8n form trigger.
   • Verify that the document is processed, chunks are created, embeddings are generated, and stored in your Pinecone index.
5. Test Chatbot Interaction:
   • Use the "When chat message received" trigger to simulate a user query.
   • Observe the chatbot's response, which should leverage the document data stored in Pinecone and the Gemini LLM.
6. Customize (Optional):
   • Adjust the "Recursive Character Text Splitter" parameters (chunk size, overlap) as needed for your documents.
   • Modify the "Information Extractor" schema to define the specific data you want to extract.
   • Implement the "Gmail" node for notifications or approvals.
   • Utilize the "HTTP Request" and "Code" nodes for additional integrations or custom logic.
   • Adjust the "If" and "Loop Over Items" nodes for more complex conditional flows or batch processing.