n8n Workflow: Extract and Structure Thai Documents with Typhoon OCR and Llama 3.1 to Google Sheets

This n8n workflow automates the process of extracting text from a specified document, processing it with an OCR engine (presumably Typhoon OCR, based on the directory name), structuring the extracted data using a large language model (Llama 3.1), and finally writing the structured data to a Google Sheet.

What it does

This workflow performs the following steps:

1. Manual Trigger: The workflow is initiated manually, allowing for on-demand processing.
2. Read/Write Files from Disk: Reads a document file from the local disk. This node is configured to read binary data, implying it's designed to handle image or PDF files for OCR.
3. Execute Command: Executes a shell command. Based on the context, this is likely where an external OCR tool (like Typhoon OCR) would be invoked to process the document and extract text.
4. Code: Processes the output from the OCR step using custom JavaScript code. This step likely extracts the relevant text from the OCR output and prepares it for the LLM.
5. Basic LLM Chain: Utilizes a LangChain "Basic LLM Chain" to process the extracted text. This chain is configured to use an "OpenRouter Chat Model".
6. OpenRouter Chat Model: Interacts with a large language model via the OpenRouter service. This model (likely Llama 3.1, based on the directory name) is responsible for structuring the raw text into a desired format (e.g., JSON, key-value pairs) suitable for a spreadsheet.
7. Google Sheets: Appends the structured data received from the LLM to a specified Google Sheet.
8. Sticky Note: A sticky note is present in the workflow, likely for documentation or temporary notes.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running n8n instance.
• Google Account: With access to Google Sheets.
• Google Sheets Credential: Configured in n8n to allow access to your Google Sheets.
• OpenRouter Account: An API key for the OpenRouter service.
• OpenRouter Chat Model Credential: Configured in n8n for the OpenRouter service.
• External OCR Tool (e.g., Typhoon OCR): This workflow assumes an external command-line OCR tool is installed and accessible on the server hosting n8n, as indicated by the "Execute Command" node. The specific command and its parameters will need to be configured.
• Document File: The input document file (e.g., image, PDF) that needs to be processed.

Setup/Usage

1. Import the Workflow: Download the provided JSON and import it into your n8n instance.
2. Configure Credentials:
   • Set up your Google Sheets credential in n8n.
   • Set up your OpenRouter Chat Model credential in n8n using your OpenRouter API key.
3. Configure "Read/Write Files from Disk" Node:
   • Specify the path to the document file you want to process.
4. Configure "Execute Command" Node:
   • Update the "Command" field with the actual command to run your OCR tool (e.g., typhoon-ocr --input {{ $node["Read/Write Files from Disk"].binary.data.fileName }} --output output.txt).
   • Ensure the OCR tool is installed and accessible from your n8n server.
5. Configure "Code" Node:
   • Review and adjust the JavaScript code to correctly parse the output from your specific OCR tool and extract the relevant text for the LLM.
6. Configure "Basic LLM Chain" and "OpenRouter Chat Model" Nodes:
   • Select your OpenRouter credential.
   • Adjust the model and prompt as needed to guide the LLM in structuring the extracted Thai document data into the desired format for your Google Sheet.
7. Configure "Google Sheets" Node:
   • Select your Google Sheets credential.
   • Specify the "Spreadsheet ID" and "Sheet Name" where the data should be written.
   • Ensure the "Operation" is set to "Append Row" or a suitable alternative, and map the LLM's output fields to the correct columns in your sheet.
8. Execute the Workflow: Click "Execute Workflow" on the "Manual Trigger" node to run the workflow.