(...and in two days I learned what CI/CD, buckets, and TailScale really are)
After qAPIx finally took shape — backend, frontend, everything neatly organized — I started wiring up the chat logic.
I designed and implemented the interaction like this:
AI asks → gets an answer → fills the block → moves on.
A simple mechanic — but the whole UX rests on it.
Then came the rest:
— Created a qAPIx-bucket in GCP to store artifacts
— Pushed everything to Git — no more keeping the project hostage in local zip archives
And finally, I decided the project had to be zen and properly managed:
✅ Created a GitHub repo
✅ Set up git pull/push from GCP using an SSH deploy key
✅ Installed Git directly on the VM
✅ The project now lives in Docker on GCP — stable and containerized
Then I tried to set up CI/CD.
It was a disaster, bro.
Spent a day and a half fighting to make it work: update the code → push → get a freshly deployed qAPIx with no fuss.
Connected Mistral and Starcoder:
— Mistral runs locally
— TailScale bridged it with GCP (yes, it’s a hack — but it works)
— Installed Ollama… and realized you can’t train models on it.
Another disaster, bro. Learned a lot, lost a few nerves.
But now qAPIx isn’t just a bunch of folders — it’s a living system with backups, pipelines, and AI.
Still… maybe I did something I’ll regret.
Why? I’ll tell you next time.