Skip to main content
PLAYER
Milos Cvetkovic
CLASS
Full Stack Engineer & Architect
SPEC
AI-Native Development
XP
13 years · 6 domains · 3 clouds
STATUS
Building at Obsidian 22

This happened at 3am.
Nobody woke up.

Milos Cvetkovic, Senior Full Stack Engineer specializing in AI-native development, TypeScript, React, and cloud architecture.

I build systems that inherit chaos and ship clarity.
Scroll to see how.

  • TypeScript
  • React
  • NestJS
  • Azure
  • Terraform
  • Claude Code
  • DDD
  • Kubernetes

PHASE 1 DISCOVERY

>

"I'm tired of 3am pages. Build something that fixes itself."

←

Interesting. What does "fix itself" mean to you?

>

Detect the error. Understand it. Open a PR with a fix.

←

Autonomous code changes need guardrails. What's the blast radius?

←

I'm thinking: budget caps, confidence thresholds, human approval...

EXTRACTED REQUIREMENTS
#1monitoring#2autonomous#3PR creation#4safety limits
QUEST LOG
✓Requirements captured
✓Constraints identified
✓Scope locked
○Architecture designed

Most bugs live in the gap between what you asked for and what you meant.

I close that gap before writing a single line of code.

PHASE 2 STRATEGY

TECH TREE

Runtime→Bun

Cold starts matter when errors are on fire

✓
Framework→Elysia

Type errors caught at compile time, not 3am

✓
AI Core→Claude Agent SDK

The part that actually thinks

✓
Monitoring→Azure Log Analytics

See everything. Miss nothing.

✓

SYNERGIES DETECTED

+Bun + Elysia = 40% smaller bundle
+Claude SDK + GitHub API = autonomous PRs
ARCHITECTURE
Azure LogsClaude AgentGitHub APIerrorsPRs

Hype fades. The right tool for the job doesn't.

I pick technologies that solve the problem, not pad my resume.

PHASE 3 EXECUTION

export class ErrorAnalyzer {
  async analyze(error: Error) {
    const context = await this.getCodeContext();
    return this.claude.diagnose(error, context);
  }
}
BUILD STATS
FILES
0%
TESTS
0%
COVERAGE
0%
TIME ELAPSED00:00:00
x0COMMIT STREAK
ACTIVITY LOG
✓src/agent/analyzer.ts — Error pattern recognition
✓src/agent/fixer.ts — Autonomous fix generation
✓src/safety/limits.ts — Budget caps, daily limits
✓tests/agent.test.ts — Unit test suite

The bottleneck was never my typing speed.

AI writes the syntax. I make the decisions that matter.

PHASE 4 THE GAUNTLET

CI/CD PIPELINE
LINT
○
TYPE CHECK
○
UNIT TESTS
○
E2E TESTS
○
SECURITY
○
BUILD
○
Achievement Unlocked
"Zero Trust, Full Send" — +500 XP

"It worked on my machine" doesn't fly here.

Six gates. Zero shortcuts. Every commit proves itself or dies trying.

PHASE 5 THE LOOP

MONITORING DASHBOARD
99.9%
UPTIME
47ms
AVG LATENCY
3
AUTO-FIXES TODAY
ERROR DETECTED
SELF-HEALING LOG
SELF-HEALING PROTOCOL ACTIVE
System diagnosed and fixed the issue autonomously

This happened at 3:14am. Nobody got paged.

The system diagnosed itself, wrote a fix, and waited for a human to approve. That's the future I build.

SESSION COMPLETE

Ideas→Architecture→Code→Production
Time: 1 conversation

This is how I work. Every time.

Follow along for more engineering deep dives.

Connect on LinkedIn

Or scroll down to see more of my work

VIEW ARCHIVE
  • AI AGENT
    RETIRED

    Self-Healing Agent

    • Claude Agent SDK
    • Bun
    • Elysia
    • Azure
    73%
    errors resolved autonomously
  • PLATFORM
    PRODUCTION

    Enterprise B2B Platform

    • React
    • Node.js
    • PostgreSQL
    • Terraform
    40%
    less complexity
  • DEVOPS
    PRODUCTION

    Nx Remote Cache Server

    • Bun
    • Elysia
    • Azure Blob Storage
    5×
    faster builds

Want to see the full architecture breakdown?

Explore All Projects

AI & Automation

Claude CodeClaude Agent SDKAI-Native Dev

Frontend

ReactNext.jsTypeScriptTailwind

Backend

Node.jsNestJSBunElysiaPostgreSQL

Cloud & DevOps

AzureAWSTerraformDockerGitHub Actions