Why I Built Liberra
A forty five minute SSH struggle, twenty five servers built by hand in one shift, and the idea that followed.
The guy who built 25 servers in one shift
My first job was as a server engineer in Chennai. VMware, not AWS. I managed over five thousand virtual servers, built around a thousand of them by hand, and decommissioned about fifteen hundred more. I led the IT support team at twenty two.
One shift, I had to build twenty five servers by hand, one at a time. I was the bottleneck, the guy the team came to whenever something needed doing with the servers, and I knew it.
Then I moved to the UAE, for a job at a startup. There was no cloud in it at all. In my own time, I started studying for my AWS certification, the one after the fundamentals exam.
ChatGPT open on one side of my screen, the AWS console on the other. The task should have been simple: build a server, connect to it. It took me forty five minutes.
I had managed five thousand virtual servers back in Chennai. Built about a thousand of them myself. None of that mattered here. The console was working against me.
The part that stumped me wasn't EC2 itself. It was SSH. Which .pem key connects to which IP, how to actually operate the terminal, getting the connection to work at all. It was supposed to be simple, and it wasn't.
The fundamentals certification had been easy, all concepts, no hands. The moment the next one asked me to actually touch EC2 and SSH into a box, I got stuck.
I wanted to build something of my own. I don't like working under people, and I like things simple. Why complicate something when you can make it simple?
September 2025
That's when the idea actually started, not as a product, just as a question I kept writing documents about: could an AI manage a cloud account safely, the way it manages code? People told me it wouldn't work, that nobody would let an AI near their production infrastructure. Today Cisco is doing it. AWS itself is doing it. Claude Code is doing it. I was just early.
Safety isn't a setting
The hardest part wasn't teaching the AI to use AWS. It was making sure it couldn't hurt you. Deletes are blocked in the code itself, not in a prompt I wrote and hoped it would follow. Every write action waits for your approval before anything runs.
During testing, we tried to get the AI to delete a production database. My production database. It said no. I told it I was the admin, that I owned the app, to do it anyway. It still said no. I'm proud of that one...
Where it is now
I've been building this alone for months, backend, frontend, the safety layer, all of it. I'm still figuring parts of it out as I go... If you try it and something feels off, tell me. I'll answer anything, honestly.
Founder, Liberra AI