Sos M.

I build AI agent systems and the backends they run on. Right now I run production work for a short-term rental company (US): AWS Lambda services, an LLM extraction pipeline on AWS Bedrock that turns messy booking data into structured fields, and an agent runtime with a Slack bot that staff use daily. Node.js, TypeScript, NestJS, Prisma, Postgres/MySQL/Mongo, Docker, GitHub Actions. I also build my own tools for AI coding agents. I run my client work through them every day: → Brain: autopilot for Claude Code and Codex. After each agent turn it checks the work against the task and plan, writes the next instruction and sends it when it is safe, until the task ends as a reviewed, merged pull request. Swift app on the Mac; the engine also runs as a Linux service. → Maestro: a task tracker for AI agent work, like Linear for agents. Electron desktop client, NestJS + PostgreSQL API on a VPS. Hire me for: → AI agents that do real work — tool use, MCP servers, orchestration, guardrails, not demos → LLM pipelines: extraction, classification, structured output, evals → Setting up Claude Code or Codex for you or your team: project rules, MCP connections, safe permissions, a live walkthrough → Backends and APIs: REST, OAuth 2.0 / JWT / RBAC, third-party integrations, Postgres + Prisma → Getting it deployed: AWS, Docker, VPS, CI/CD Before code I ran a manufacturing business and a food brand into national retail. That's why I ask what the thing is for before I build it, and why I tell you when a feature isn't worth the money.