VeriMine

Inspiration

Every small business we know in El Salvador runs on data nobody looks at. Sales live in one Excel file, inventory in another, orders arrive over WhatsApp, and payroll is a notebook. The data is there — piling up, month after month — but the owner still decides by gut feeling, because opening those files raises more questions than answers. When we sat down to build our demo dataset, we didn't have to invent the mess. We just described what we've actually seen: the same branch written nineteen different ways across one file, money stored as text with mixed currency symbols, totals and taxes blank on every row, and a summary sheet that flat-out contradicts its own detail rows. A company that "has data" like this doesn't have an asset. It has a liability it hasn't noticed yet. So the obvious move is to paste the file into a chatbot and ask. And the chatbot answers — fluently, confidently, with charts-ready numbers. Some of them are made up. There is no way to tell which ones without re-doing the analysis yourself, which defeats the point. For a business, a wrong number is worse than no number: you order stock you don't need, you chase the wrong branch's "growth", you set goals against revenue that was never real. That gap is what inspired VeriMine. These companies don't need a data science department — they need what a good data team does: someone who goes and gets the data, someone who cleans it, someone who analyzes it, and someone who checks the math before it reaches the boss. We realized a multi-agent system could be that team. But the last role — the one who checks — is the one nobody builds, and it's the one that matters most. So we built the team, and then we built the checker: every figure our agents report must come with the SQL that produced it, get re-executed against the real data, and get sealed into a hash chain, so the numbers can be audited and the data can't be quietly altered mid-analysis. Our rule became the product's name: don't trust, verify.