I began building Python applications for CSV/Excel ingestion, header detection, bulk SEO, AI generation, product cleanup, partial saving, resume logic and book-data verification.
Software built around the operator
The applications were not designed from a theoretical product brief. They came directly from the way ecommerce work actually happens: files arrive with different headers, only certain rows need processing, existing fields cannot be overwritten, long jobs must be stoppable and useful partial outputs must survive.
Those details are small from a software-demo perspective but critical in a production workflow. They are the difference between an AI experiment and a tool an operator can trust.
Book verification closed an old loop
A separate book-data verification direction connects back to the earliest KDP and book experience: ISBN, title, author, publisher and description data need to be correct before enrichment is valuable.
The goal moved beyond generating text. The system should read existing product information, verify what it can, preserve what is correct and improve only what needs work.