Building Things That Hold Up in Production

Most of what I build starts the same way: a problem that feels slightly too annoying to ignore. Then comes the sketch, the first broken version, and the quiet realization that the interesting work is still ahead.

At Heuristic Labs and GGS, that loop got faster. Interfaces had to feel intentional. APIs had to stay predictable under load. AI features had to fail gracefully instead of impressively.

PrivateGPT taught me the value of keeping sensitive work local. The project management platform taught me how far indexing, RBAC, and careful schema design go when concurrency stops being theoretical.

I still care about polish. I just care more about whether the thing still works when nobody is watching the demo.