vml/soft

Funen, Denmark · building software since 2008

Vlad Rotaru

Senior software engineer. I design, build and run back-end and full-stack systems for product teams, with an AI engineering setup I built and operate myself.

What I do

AI engineering for your team

I set up coding agents, review gates and guardrails that fit how your developers already work. We start on a real repository and end with a written playbook your team owns.

Contract development

I join your team as a senior engineer: back-end services, APIs, data models and the front end that sits on them.

Hard business logic

Pricing, catalogues, rules engines: the parts where a wrong number costs money. I like making them correct, tested and fast.

Reviews and rescues

A second pair of eyes on a codebase, an architecture or a migration that went sideways. Written findings, ranked by what matters.

Most of my client work runs on Go, TypeScript (Angular, Node) and PHP, on Google Cloud and AWS: cloud marketplaces, billing and pricing engines, and identity integrations with Microsoft Entra ID, SCIM and PingOne.

How I work

I work with AI the way a senior engineer works with a good team. On my own infrastructure I run coding agents that implement, review and test, and I direct them, check their work and own every line that ships.

For clients that means more ground covered per day, with the same standard: written plans, tests before merge, and code you can read after I leave.

Client code and data never go to a tool you have not approved.

The AI engineering setup I run

Not a chat window. A working system of agents across several machines, built and operated by me, that I use on client work every day.

orchestration

Multi-model build pipeline

One model plans and reviews, others implement in parallel. Every change is checked against a written plan and its tests before it merges.

coordination

Agent message bus and board

Agents on different machines hand work to each other over a message bus and record decisions on a shared board.

machine agents

An agent per machine

Each machine has its own on-demand agent with its own memory, that can split work into parallel sessions and report back.

guardrails

Unattended jobs with a human gate

Scheduled agent jobs run with explicit permission lists, fail loudly on anything denied, and pause for my approval on the phone before risky steps.

tooling

Shared skills and tools

My own plugin catalogue ships skills, tool servers and safety hooks to every machine, so all agents work the same way.

control

Phone-first operations

A native iPhone app and assistant for status, notifications and approvals, on a self-hosted platform with its own git hosting, CI and local models.

Working on something?

Write a few lines about the project and the timeline. I reply in English or Danish.

vlad.rotaru@vmlsoft.com