AI infrastructure engineer in Germany

What an AI infrastructure engineer does

An AI infrastructure engineer runs a company’s models and agent software, and the servers under them. The work covers the server or cloud region, the serving software that answers requests, the model, the agent harness and the rules for who may call what. An AI engineer builds the applications on top.

On your own servers, prompts and documents stay on machines you choose, and the model version changes only when your team decides.

Open weights and open harnesses, checked in September 2026

Both halves of the stack are available under open licenses, the model and the harness around it.

DeepSeek released a preview of DeepSeek V4 on April 24, 2026, in two sizes, V4-Pro and V4-Flash (announcement). The weights of DeepSeek-V4.1-Flash are published under the MIT license, with instructions for vLLM and SGLang on the model page. DeepSeek’s own API accepts the OpenAI and the Anthropic format (API docs), so software written for either format reaches it after a change of the base URL. The licenses of Qwen, Mistral and gpt-oss are compared on Self-hosted LLMs such as DeepSeek.

DeepSeek Harness is an open-source agent harness from DeepSeek under the MIT license. It runs on Node.js, and everything in it is a plugin (GitHub). InfoQ reported on August 20, 2026, that a change in its configuration file moves the model between remote API providers and local runtime servers (InfoQ). The project is a developer preview, and its maintainers announce breaking changes.

The model and the harness are parts your team can exchange. The engineer you hire keeps the rules and the access limits in place while they change.

The job description, layer by layer

The table lists what an AI infrastructure engineer runs and the part of my work each layer draws on. Hold it next to your own job description.

Layer	What the engineer does	Where my experience comes from
Hardware	Sizes and runs servers in your rack or at a hosting provider	Servers, switches and firewalls I installed in data centers in Germany and the United States; 40 load-balanced servers for a web community
Cloud and failover	Runs the service in a cloud region with a second site	AWS, Azure and Google Cloud; a node operation moved from Azure to another cloud with failover
Services that stay up	Keeps software running for users who depend on it	Blockchain nodes and an oracle network in several data centers, on bare metal and in the cloud, set up as scripts and Docker containers
Serving the model	Installs serving software such as vLLM, with a token per application and a log of the requests	A setup I plan for companies; the server work above carries over
Agent harness	Runs Claude, Codex or an open harness under the company’s rules	Claude and Codex under one rule file that the Gemini CLI reads too, with rules and memory outside the model
Access and limits	Decides what an agent may read and change	A read-only database role, scoped tokens, a protected production branch and hooks with a test matrix
Data	Connects the model to the company’s data through APIs	A client platform, rebuilt with Claude and Codex, has more than 300 RESTful API endpoints and more than 8,000 commits

Who I am

I am Michael Wutzke, an IT expert for data platforms in Frankfurt. I have spent more than twenty years from code to CIO in finance, e-commerce and real estate, with cloud platforms and data centers. Today AI agents speed up how I build and transform data platforms.

In client work I rebuild a platform with Claude and Codex under one rule file. I teach at Claude Hacker House. I have run blockchain nodes in several data centers. I hold the AWS Solutions Architect Associate certificate.