Spent 2 hours reconciling 3 Excel sheets today. Built an AI tool to do it in 30 seconds. Looking for beta testers.

I work in HFT where a manual Excel error isn't just a typo—it's a compliance fine. We spent months fighting with:

• Macros: Breaking every time a broker added a column.
• Vision-based AI: Way too slow/unreliable for multi-sheet reconciliation.

We decided to build a custom execution engine that hooks into the Excel object model directly. Instead of "looking" at pixels, it maps the sheet structure dynamically. It turned our 45-minute daily reconciliation into a 30-second automated run.

The Tech: It’s built as an MCP (Model Context Protocol) server. You give an LLM the "intent," and the engine handles the cross-sheet lookups and normalization without the AI having to "guess" where the data is.

Question for the community: For those of you doing high-stakes reconciliations—what’s your biggest "failure point" when trying to automate? Is it the file structure changes, or the cross-referencing logic?

I’m refining the engine and would love to see if this logic holds up against other messy sheet structures. Happy to share what we've built if anyone wants to stress-test it.