A data engineer describes building a PDF invoice data extraction pipeline using Snowflake's AI_EXTRACT SQL function and Cortex Code (an AI coding assistant integrated into Snowsight). The project involved extracting EV charging session data from three evolving PDF invoice layouts (v1, v2, v3) in German, normalizing the results into a harmonized Snowflake table, and setting up stored procedures for initial and incremental loads. The author shares honest reflections on the human-AI collaboration model: the AI excelled at code generation, pattern propagation, and documentation, but required human oversight for schema design, data quality validation, and correcting subtle bugs like a decimal separator normalization error.
Table of contents
ContextMeet my AI Coding Buddy as virtual teammateExploring AI_EXTRACT with Cortex CodeGet Harald Erb’s stories in your inboxFrom exploration to robust data extraction and ingestionMy AI Coding Buddy — always on and ready to solve more problemsConclusion: It’s more fun to compute — with my AI Coding BuddySort: