This project doesn't solve a burning problem for me, but it's more of a curiosity-driven experiment asking one question: with LLMs and fuzzing, how close can I get to a 100% Excel-compatible spreadsheet engine?
After working on this for a couple of months, it seems to be pretty close now (and in some cases, more accurate than Excel). Recent fuzzing runs have been unable to find discrepancies with Excel, but I think that there is still a long tail of cases out there. Nonetheless, visi has worked well so far for casual agentic use cases, e.g. modeling the minimum spend to break even on a credit card annual fee/running a DCF model for a company. To view the resulting workbook, I've been using this OOXML viewer (HN link): https://news.ycombinator.com/item?id=48436863
Note: the name "visi" is a nod to VisiCalc. It is also easy to type.
Comments URL: [https://news.ycombinator.com/item?id=49929194](https://news.ycombinator.com/item?id=49929194)
Points: 1