How I Built a Zero-Upload JSON, Bank PDF & Invoice to Excel Workbench in the Browser
Most online file converters follow the same pattern: upload your file to a remote server, wait in a queue, and hope your financial data actually gets deleted. When building JSONtoSheet , I wanted to see how much of a real bookkeeping and developer spreadsheet workflow could run 100% inside local browser memory. 1. Flattening Nested JSON into Native BIFF8 .xls and .xlsx Enterprise ERP tools still…
Building a zero-upload JSON, bank PDF, and invoice to Excel workbench in the browser is a unique approach to financial data conversion. Most online file converters rely on uploading files to remote servers, waiting in a queue, and hoping the data gets deleted safely. However, when creating JSONtoSheet, the goal was to create a fully local browser-based workflow.
One challenge was flattening nested JSON data into native BIFF8 .xls or .xlsx formats. Legacy ERP tools often require these older formats, but SheetJS can handle this using recursive dot-notation flattening. Developers can paste raw API responses, adjust columns live, and export clean workbooks without any data leaving the browser tab.
Another hurdle was parsing bank statement PDFs and invoices locally. Bank statements are notoriously difficult to parse because multi-line descriptions often wrap across rows. The solution was to extract text coordinates locally, reconstruct rows by Y-axis tolerance, and run automatic balance and line-item checks. This way, missing rows are flagged immediately, ensuring data accuracy.
Lastly, there's a converter for directly mapping CSVs to QuickBooks .qfx files. Many small banks only export generic CSVs, while QuickBooks Desktop expects Web Connect .qfx (SGML/OFX) format. By mapping Date, Description, and Debit/Credit columns client-side, the CSV to QFX Converter creates valid QFX files without any server involvement.
Throughout the development process, it became clear that a hybrid approach, combining deterministic local parsers with AI for challenging cases, offers the best results. Running deterministic parsers first can instantly convert 80% of cases without any cost. Saving AI resources for skewed scans or thermal receipts keeps the project's unit economics sustainable.
Most importantly, privacy is a significant factor for bookkeepers and developers. Keeping financial data local is a major concern, so testing the workbench at jsontosheet.com is encouraged to see what breaks and how it can be improved.
Rewritten by urgent.news; may contain errors — check the original.
Source: dev.to