AI Invoice Data Extraction Agent with LlamaParse & OpenAI

Unlock Automated Invoice Processing with this AI Agent

This AI Agent streamlines your accounts payable by automatically processing PDF invoices received via email. It leverages LlamaParse for robust PDF-to-markdown conversion, intelligently handling complex layouts and tables often missed by standard parsers. OpenAI's GPT model then extracts key invoice details (dates, numbers, supplier info, line items, totals) according to a defined structure, ready for direct input into your financial systems or spreadsheets.

Key Features & Benefits

• Advanced PDF Parsing: Utilizes LlamaParse to accurately convert complex PDFs, including those with tables and figures, into machine-readable markdown. This AI-driven ability ensures comprehensive data capture from challenging document structures.
• Intelligent Data Extraction: Employs OpenAI (GPT-3.5-turbo or similar) to precisely extract predefined fields like invoice numbers, dates, supplier details, customer information, line items, and various totals. This is the core 'data analysis' ability of the agent.
• Structured Output: Guarantees AI-extracted data is formatted correctly into JSON using a structured output parser, enabling seamless integration with other systems.
• Automated Data Logging: Directly appends extracted invoice information to a Google Sheet for easy reconciliation, reporting, and record-keeping.
• Email Integration & Monitoring: Actively monitors a Gmail inbox for new emails with PDF attachments matching specified criteria.
• Duplicate Prevention: Automatically labels processed emails in Gmail to avoid redundant operations and maintain data integrity.
• Resilient Processing: Includes checks for LlamaParse job completion and wait steps to manage service limits, ensuring reliable operation.

• An n8n instance (Cloud or self-hosted).
• OpenAI API Key with access to a model like gpt-3.5-turbo-1106 or newer.
• LlamaCloud API Key (LlamaParse is a service within LlamaCloud, free tier available at time of writing).
• Gmail credentials (OAuth2) for reading emails and adding labels.
• Google Sheets credentials (OAuth2) for writing extracted data.

1. Download the n8n workflow JSON file.
2. Import the workflow into your n8n instance.
3. Configure the 'Receiving Invoices' (Gmail Trigger) node: set your email filters (e.g., specific sender, subject line has:attachment) and connect your Gmail account via OAuth2.
4. In your Gmail account, create a label named "invoice synced". This label is used to prevent reprocessing. If you use a different label name, update it in the 'Should Process Email?' node and the 'Add "invoice synced" Label' node.
5. Configure the 'Upload to LlamaParse', 'Get Processing Status', and 'Get Parsed Invoice Data' (HTTP Request) nodes. In their credentials section (Header Auth), add your LlamaParse/LlamaCloud API Key (usually a Bearer Token).
6. In the 'OpenAI Model' node, select or add your OpenAI API credentials and ensure a suitable model (e.g., gpt-3.5-turbo-1106) is selected.
7. Review the 'Structured Output Parser' node. The jsonSchema defines what data points will be extracted. You can customize this schema if your invoice data requirements differ. The sticky note near this node in the workflow offers guidance.
8. Examine the prompt in the 'Apply Data Extraction Rules' (LLM Chain) node. This prompt instructs the AI on how to extract information from the markdown content provided by LlamaParse. Adjust if necessary for specific invoice layouts or data points.
9. Configure the 'Append to Reconciliation Sheet' (Google Sheets) node: connect your Google Sheets account, specify the Spreadsheet ID (from the URL) and the Sheet Name (e.g., 'Sheet1' or 'gid=0'). Ensure the columns in your Google Sheet match the fields defined in the jsonSchema or adjust the mapping in the 'Map Output' node if needed. The sticky note near this node in the workflow offers guidance.
10. Activate the workflow. New emails matching your criteria should now be processed automatically.