Dynamic AI Data Extraction from PDFs to Airtable - n8n Workflow

Automation Specialists needing a flexible n8n template for database synchronization and AI enrichment.
Data analysts who use Airtable as a central source and require automated PDF parsing.

• Developers looking for a complex example of conditional logic and dynamic prompt generation in an n8n node flow.

This comprehensive n8n workflow solves the challenge of automating data population in a structured database (Airtable) based on unstructured input (PDF files) and flexible requirements (dynamic prompts). Instead of hard-coding extraction fields, this n8n template reads the extraction instructions (prompts) directly from the Airtable column descriptions.

It operates using a reactive approach: the workflow is activated by an n8n trigger (Airtable Webhook) whenever a row is updated (new file uploaded) or a field definition is changed (new prompt defined). It smartly routes the execution to either update a single row (for performance) or recalculate an entire field across all rows (for schema changes). This design makes the n8n workflow highly efficient and adaptable to evolving data needs, proving the power of a highly customized n8n solution.

The n8n workflow begins with the Airtable Webhook n8n trigger listening for table changes (row updates, field creation/updates).

1. Schema Retrieval: Upon triggering, the workflow first retrieves the table schema using an Airtable n8n node to identify all field descriptions, which serve as the dynamic prompts.


2. Event Parsing and Routing: A Code n8n node parses the event payload to determine the exact event type (row.updated vs. field.created/field.updated). The Switch n8n node then routes the flow accordingly.


3. Targeted Row Update (Input File Change): If a row is updated, the n8n workflow identifies that specific row and uses a Code n8n node (Fields to Update) to determine only the fields that are currently missing data. It iterates through these fields (Loop Over Items1).


4. Full Field Recalculation (Prompt Change): If a field description (prompt) is changed or created, the n8n workflow fetches all relevant rows containing a file input. It then iterates through every applicable row (Loop Over Items).


5. AI Extraction: For both branches, the n8n node structure is similar: it uses an HTTP Request n8n node to download the PDF, an Extract from File n8n node to convert the PDF content to text, and a powerful LLM n8n node (Generate Field Value) to run the dynamic prompt against the extracted text. The prompt includes instructions on the required output format, ensuring quality results.


6. Database Update: The extracted data is collected using a Set n8n node and finally written back to the specific Airtable record using an Airtable n8n node, completing the entire automated n8n workflow.

To deploy this expert n8n workflow, follow these steps:

1. Import: Import this n8n JSON file into your n8n instance.


2. Credentials: You need two sets of credentials:

1. Webhook Setup (Crucial): Airtable webhooks must be manually configured (or use the provided setup sub-flow triggered by When clicking ‘Test workflow’):

Airtable Webhook (n8n trigger): The initial n8n trigger, listening for POST requests from Airtable indicating changes in rows or fields. Crucial for the reactive nature of this n8n workflow.
Get Table Schema (Airtable n8n node): Fetches the full metadata of the Airtable base and table, specifically gathering the column descriptions used as dynamic prompts.
Parse Event (Code n8n node): Analyzes the raw webhook payload to distill key information: base ID, table ID, the exact event type, and affected row/field IDs. This is core logic for the subsequent conditional routing in the n8n workflow.
Event Type (Switch n8n node): Directs the flow based on whether a specific row was updated or a field schema was modified. A classic n8n node used for flow control.
Extract from File (n8n node): Downloads the file (usually a PDF, via a preceding HTTP Request) and extracts its text contents, preparing the context for the LLM. Essential for PDF processing within the n8n workflow.
Generate Field Value (LLM n8n node): This Langchain n8n node takes the extracted PDF text and the dynamic prompt (from the field description) to generate the required data point. It leverages OpenAI credentials configured for the n8n template.

• Split in Batches (n8n node): Used to iterate over multiple records efficiently, ensuring each row is processed sequentially before updating Airtable. This n8n node manages looping in the flow.