Fifteen years running commercial operations across the Baltics, including a country P&L. Now I build the software myself — and everything below is live, or was, rather than a prototype in a folder.
Real users, real payment infrastructure, real consequences when it breaks.
A rep talks about the visit; the app extracts the doctor, the product, the sentiment and the follow-up, and writes the record. Built end to end: Postgres with row-level security, Stripe billing with VAT handled, fuzzy matching so a spoken name attaches to the right existing record instead of creating a duplicate, territory mapping, an AI next-visit brief that invalidates itself when the underlying visits change. Own domain, full GDPR set including a DPIA, ten indexed articles.
repflow.northlightpress.net →Thirty checks to run against your own Supabase-backed app before you take money through it, each one with the exact SQL or browser test. Written after finding a SECURITY DEFINER function on my own production database quietly carrying Postgres's default PUBLIC execute grant — which the build tool's own summary said had not happened.
FIFO and HIFO realised gains, short and long split, loss carry-forward, crypto-to-crypto swaps, CSV export. Entirely client-side — no account, nothing uploaded. The calculation engine was independently validated before release.
Open the calculator →Matches every receipt against every line on a bank statement and produces the list of what is unaccounted for, then hands a clean folder to any accountant. Built because I was doing it by hand every month. Deliberately not an accounting system — it does the handover and stays out of the way.
Financial models and trackers built in Excel and Google Sheets — budgeting, net worth and FIRE projection, crypto tax-lot tracking, a freelancer finance system in EU and US variants. Storefront, payment processing, VAT handling and delivery all configured and working.
The storefront →Internal tooling, so there is no public link — described honestly rather than dressed up.
A working pipeline that takes a real company's material and produces scheduled social content on a routine, in a named vertical, for an operating business.
Scans multiple job boards on a schedule, filters against a specific set of criteria, and reports only what genuinely qualifies — including saying "nothing today", which is the harder behaviour to build.
Fifteen years in commercial operations, which is why the automation tends to target the right step.