What is your goal?
Step 1: The “Digital Watchman”
First, I set up a monitoring system that constantly watches a specific folder in my cloud storage. The moment a new file—containing both text and images—is uploaded, my automation “wakes up” and grabs the file’s ID so it knows exactly which document to work on.
Step 2: The “File Fetcher”
Once the file is detected, I have the system perform a “Download” action. This step is crucial because it converts the file from a simple link into a raw “binary buffer”—essentially the digital DNA of the document—that an AI can actually “see” and analyze.
Step 3: The “AI Brain” Analysis
This is where the magic happens. I send that raw file data to a high-powered AI model. I’ve written specific JSON instructions to tell the AI: “Look at this document and extract all the key findings and defects into a structured list”. This replaces the 20 minutes I would usually spend reading the report myself.
Step 4: The “Secure Vault” Storage
Finally, I take the structured text the AI produced and tell my database to “Create a New Row”. The goal is for those findings to appear instantly in my findings table so I have a clean, searchable record of every document processed.
What is the problem?
I can’t get the standard “File Data” input field to appear in the module’s interface. To get around this, I turned on the Map toggle to manually write the JSON structure.
Even though I’ve tried to format the JSON exactly as the API expects, I keep getting this error:
BundleValidationError: Validation failed for 1 parameter(s). Array of objects expected in parameter ‘contents’.
What have you tried so far?
I successfully ran the preceding file download step, so I know I have a valid binary buffer (the yellow “Data” variable).
I removed the double quotes around the Data variable so it stays as raw data rather than a text string.
My current JSON structure looks like this:
[{“role”: “user”, “parts”: [{“fileData”: {“mimeType”: “application/pdf”, “data”: {{data}} }}, {“text”: “Extract details…”}]}].








