@szymon_dybczak  it worked 

The code reads raw text content extracted from PDF files stored in the Delta table pdf.default.raw_pdf_text

Databricks ai_query function, which internally calls a large foundation model (databricks-meta-llama-3-3-70b-instruct) to process the raw text

The transformed data, which now includes the extracted structured information alongside any original fields, is saved to a new Delta table pdf.default.parsed_pdf_data