A freelancer found their paid bookkeeping tool too slow and lacking in essential reporting. They decided to build their own solution using Python, resulting in a small codebase with extensive testing. This custom system includes an invoice scanner with human oversight for data extraction. The process automates monthly bookkeeping tasks, generating an Excel report for their accountant. While tempted by AI, the freelancer identified only one step needing AI: extracting data from varied supplier PDFs. The rest of the process relies on robust code for arithmetic and validation. The system emphasizes a human-in-the-loop for critical decision points and data approval. Once approved, PDFs are renamed to serve as immutable records, eliminating the need for a database. The system's design prioritizes immutability and traceability, with all totals in the final report being formulas. This approach ensures easy verification by the accountant. The freelancer now maintains the system, but gains full control, reduced costs, and an improved workflow. Key takeaways include mapping workflows, using AI sparingly, coding rules, human oversight, and ensuring traceability.