Treating LLM prompts like code: a regression catalog for AI failuresA field note on turning prompt-engineering folklore into structured, regression-tested artifacts. A few months ago, an LLM call in the system I work on confidently paired a refund with a beneficiary12EMay 20 · 10 min readreconpe.hashnode.dev
What I learned trying to let AI handle financial reconciliation - and why I stoppedA pragmatic note on where large language models earn their keep in financial reconciliation, and the unsexy engine I had to build to keep them honest. For a few months in 2025, I tried to let a langMay 11 · 9 min readreconpe.hashnode.dev
We use AI everywhere in our reconciliation system. We don’t let it pick the matches.A field guide to where large language models genuinely help in financial reconciliation — and the deterministic engine we built to keep them honest. If you read fintech marketing in 2026, you would May 5 · 8 min readreconpe.hashnode.dev
What we learned building reconciliation that doesn’t forgetMost recon tools treat every month-end as an island. We tried to change that. Here’s what worked, what broke, and the deliberately boring math underneath. In February, one of our finance users flaggMay 1 · 6 min readreconpe.hashnode.dev
