Jorge Filho
I take a problem nobody has written down yet and hand back a system running in production. And still running two years later. I work above the code.
Systems delivered.
Not demos. Systems that carry real money, day after day, and are still running.
Logs into a card processor, matches each agent by name with an override table for homonyms, and alarms after two failures. 353 reconciled in a month, one gap recovered.
One isolated stack per client, with its own secrets and an empty database, up in about ten minutes. Five systems now run on it.
Row-level proof that source and destination matched exactly. Found a live exposure on the way in, customer ID scans at public URLs, and closed it.
Face recognition that keeps only the math embedding, never a photo. A rule I set for Brazil's data-protection law before any client asked.
# 06:00 sweep $ curl -s /health { "ok": true, "rows": 43, "lost": 0 }
Nothing is done until I've seen it work.
A stale proxy once returned a clean 200 and dropped a full day of data. After that, my systems check their own answer, not the messenger.
# every deploy auth server-side ✓ creds never in browser ✓ ports 127.0.0.1 only ✓ secret per stack ✓
Open source.
Client code stays private. These are the reusable parts, four angles on how I build.