The short version
I fix the systems everyone else gave up on.
Then I make them better than they were before the problems started.
You know that codebase? The one with the "temporary" workaround from 2017 that somehow became load-bearing? The one where three developers quit rather than touch the payment module? The one everyone says needs a "complete rewrite" but nobody has two years to spare?
That's my favorite kind of project.
I've spent a decade inside systems like that. Not just surviving them—transforming them. Untangling dependencies. Introducing tests where there were none. Building architecture that makes the next change possible instead of terrifying.
But here's what changed: I got tired of being the only one who could fix things. So I started building AI that works the way I do. My self-healing agent monitored production 24/7, diagnosed errors, and opened PRs with fixes—without waking anyone up.
It's not about replacing engineers. It's about giving them superpowers.
What does Milos Cvetkovic build?
Two kinds of system. The first is the legacy platform nobody wants to touch: I introduce boundaries one module at a time, so every pull request ships value while the architecture improves underneath it. The second is AI that repairs what breaks — a self-healing agent, since retired, that watched production, diagnosed errors and opened pull requests a person reviewed and merged.
What stack does Milos Cvetkovic work in?
TypeScript end to end. React and Next.js on the front, NestJS, Node and Bun with Elysia behind it, PostgreSQL for state. Infrastructure is Azure — Container Apps, Blob Storage, Log Analytics — described in Terraform and shipped through GitHub Actions with Nx. The AI work runs on the Claude Agent SDK. Clean Architecture and domain-driven design are the habits underneath all of it.
What was Milos Cvetkovic's self-healing agent, exactly?
A service on Azure that watched production logs, read the codebase, and when something broke, diagnosed the error and opened a pull request with a fix. It ran on Bun and Elysia with the Claude Agent SDK, and it shipped with limits: a daily cap, a budget cap, confidence thresholds, capped CI retries and a kill switch. A person reviewed and merged every fix.
Quick facts
The longer version
AI-Native Engineer
"Built an AI agent that fixed production bugs while I slept"
Combining a decade of battle scars with cutting-edge AI. My self-healing agent diagnosed production errors and opened pull requests with the fixes, which I reviewed and merged: 73% errors resolved autonomously, no 3am pages.
JavaScript Tech Lead
"Fewer bug reports after architecture overhaul"
Inherited a codebase where "temporary fixes" had calcified into permanent nightmares. Introduced Clean Architecture. Watched bug reports drop. Trained the next generation of leads.
Full-Stack Developer → Tech Lead
"First microservices migration, first cloud deployment, first gray hairs"
The years that taught me everything breaks eventually—and how to build systems that break gracefully. Migrated monoliths to microservices. Learned why "it works on my machine" is a confession, not an excuse.
Frontend Developer
"Survived jQuery spaghetti and the AngularJS-to-Angular migration"
Where the obsession began. Discovered that my favorite problems are the ones everyone says can't be solved. Still true.
What I believe
Shipping beats perfection
A working feature today beats a perfect feature next quarter. I've seen too many "almost done" projects die in committee. Ship it, measure it, improve it.
Automation is self-respect
If I'm doing the same task twice, I'm building a tool. Life is too short for manual deployments and copy-paste workflows. Robots should do robot work.
Clarity over cleverness
The cleverest code I've ever written was also the most expensive to maintain. Now I write code for the tired developer at 2am who just needs to understand what's happening.
Credentials
Let's connect
I share engineering insights, open-source work, and lessons learned from the trenches. Follow along or drop me a message.
Last updated .