Structured Document Data Extraction Pipeline with GPT-4 and PostgreSQL - n8n Workflow

Data Engineers needing automated document ingestion pipelines.
Finance or Accounting teams automating invoice processing.
Businesses requiring reliable extraction from contracts or reports.
n8n power users looking for advanced AI and database integration n8n templates.

Manual data extraction from unstructured documents like invoices, contracts, and reports is slow, error-prone, and resource-intensive. This comprehensive n8n workflow provides an end-to-end solution for turning incoming files into clean, validated, structured data.

This n8n automation starts with a local file n8n trigger, uses a specialized PDF n8n node for document processing, and leverages the superior extraction capabilities of GPT-4, forcing it to return a reliable JSON object. A custom Code n8n node then ensures data quality before the information is routed based on document type. Whether you are building new n8n templates or optimizing existing processes, this n8n workflow demonstrates best practices for integrating advanced AI and robust database operations.

This n8n workflow operates entirely automatically once deployed, starting with a powerful n8n trigger and flowing through several critical steps:

1. File Monitoring (Watch Folder): The process is initiated by the Local File n8n trigger, which monitors a specified local path (/documents/incoming) for newly created files.


2. Document Parsing (PDF Vector): When a file is detected, the document path is passed to the PDF Vector n8n node, which intelligently parses the PDF content using its built-in Large Language Model (LLM) capabilities to ensure maximum text extraction accuracy.


3. AI Extraction (GPT-4): The extracted content is sent to the OpenAI n8n node, configured to use GPT-4. The prompt strictly asks for key structured information (like Document Type, Dates, Amounts, Parties) and forces the output into a JSON object format.


4. Validation and Cleaning (Code Node): The extracted JSON data is then processed by a custom Code n8n node. This essential step validates document types, ensures dates are correctly formatted (ISO 8601), cleans monetary values by stripping non-numeric characters, and validates email addresses, preparing the data for the database.


5. Routing (Switch Node): A Switch n8n node examines the validated documentType. If it matches 'invoice', the workflow follows Route 1. All other types follow the default route (Route 2).


6. Database Storage (PostgreSQL): Route 1 uses a PostgreSQL n8n node to insert specific fields into the invoices table. Route 2 uses a PostgreSQL n8n node to store general document metadata and raw content in the documents table.


7. Final Export (CSV): Both database branches converge to the Write Binary File n8n node, which compiles all processed items and exports a final summary CSV file to a local path.

To deploy this comprehensive n8n workflow:

1. Import the n8n templates: Copy the provided JSON and paste it into your n8n instance using the 'New' -> 'Import from JSON' option.


2. Credentials Setup: Configure the necessary credentials for the following n8n nodes:

1. Local File Trigger: Ensure the path configured in the Local File n8n trigger (/documents/incoming) exists and is accessible by the n8n host system.


2. PDF Vector Node: This node may require specific credentials or setup depending on your PDF Vector configuration.


3. Database Preparation: Ensure the target PostgreSQL tables (invoices and documents) are created and the column names match those specified in the PostgreSQL n8n nodes.


4. Activate: Save and activate the n8n workflow. The n8n trigger will now be watching the specified folder.

Watch Folder (Local File Trigger):
Function: Serves as the initial n8n trigger, starting the automation when new files are dropped into the specified folder.
Key Configuration: Path set to /documents/incoming; Events set to file:created.
PDF Vector - Parse Document (PDF Vector n8n node):
Function: Converts the binary document into readable text content, utilizing LLM capabilities for superior parsing.
Key Configuration: Resource: document; Operation: parse; Use LLM: always.
Extract Structured Data (OpenAI n8n node):
Function: Uses GPT-4 to analyze the extracted text and structure the key data points into a standardized JSON format.
Key Configuration: Model: gpt-4; Options: Response Format Type set to json_object to ensure structured output.
Validate & Clean Data (Code n8n node):
Function: A crucial customization point in the n8n workflow that validates and standardizes the extracted data (e.g., date formats, cleaning currency symbols) before database ingestion.
Key Configuration: Custom JavaScript logic for data cleaning and validation.
Route by Document Type (Switch n8n node):
Function: Directs the data flow based on the documentType extracted by the AI, separating invoices from general documents.
Key Configuration: Condition checks if documentType equals invoice.
Store Invoice Data / Store Other Documents (PostgreSQL n8n node):
Function: Persists the processed, validated data into the designated PostgreSQL tables.
Key Configuration: Operation: insert; Tables: invoices or documents.
Export to CSV (Write Binary File n8n node):
Function: Aggregates the results from the database insertions and generates a local CSV summary file for auditing or further processing outside this n8n workflow.
* Key Configuration: File Content uses a custom expression to convert all incoming items to CSV format.