Build · AI development

I build AI systems that hold up in production.

Many AI projects work in the demo and stall before daily use. I design the system, connect it to your data and systems and take it into production myself. You get a defined outcome at a fixed price, including the handover to your team.

Fixed price Defined scope Handover to your team

Starting point

Where AI projects usually get stuck before production.

Quality cannot be measured

Nobody can say whether a change makes the system better or worse. Without test cases (evals), every adjustment becomes a risk.

Agents can do too much

AI agents often get more access than their task needs. What happens when third-party content leads them astray was never checked.

The connection is missing

The AI is not properly connected to the systems your people work in. Add legacy systems, patchy data and unclear ownership.

What I build

For two starting points.

Whether you develop your own AI or roll out existing tools, the goal in both cases is a system that holds up in daily use.

You build your own AI

Prototype, MVP, AI platform or agent system. I take responsibility for the architecture, measurable quality (evals) and the security review before go-live. I build the parts that decide whether it runs myself: the connection to your systems (MCP integrations), continuous quality measurement and the safeguards.

You roll out existing tools

For tools like Copilot, ChatGPT or Langdock, I set up the rules, integrations and training that make them part of daily work. For smaller companies there is a dedicated format for this, the AI-native setup.

See the AI-native setup →

Experience

I only advise on what I can build and verify myself.

“Simon designed the architecture around our compliance agent KAIA and brought it to paying customers. He was instrumental in establishing evals and made the quality of our AI measurable. Simon takes on far-reaching responsibility, brings the team along and ensures the highest level of security and reliability, even on the most complex tasks.”

AP

Alexander Prams

CTO & Co-Founder, Kertos

Translated from German.

Kertos · Architecture

AI platform from scratch

Architecture and agent system for a compliance platform, from the first line of code to LLM observability in production. MCP integrations, evaluation pipelines, Datadog.

My own company and the AI workflows of a research institute run on systems I built and operate myself.

How it works

From the goal to the handover.

01

Agree goal and scope Before we start

We clarify what should change in a measurable way and how you will recognise success. This gives us a defined scope and a fixed price.

02

Build Independently

I build the system with my own tools and connect it to your data and systems. Test cases that make quality measurable (evals) run from the start.

03

Take it into production Go-live

I make visible what the system does in production (LLM observability) and check before go-live what happens when users or third-party content lead it astray.

04

Hand over Completion

Your team gets documentation, the eval suite and code reviews that explain the reasoning. With these, it can run and develop the system itself.

For a first use case there is the use-case lighthouse format: six weeks to a running system.

Where my work ends

Legal questions are for lawyers.

Whether a system complies with the GDPR or the EU AI Act is not mine to judge. I provide the technical facts for it: which data flows where and which vendors are involved.

The fixed price covers the agreed scope.

Anything new that comes up along the way, we discuss before I build it.

Your team stays in control.

Code and architecture belong to you. I build so that your team can keep building without me.

Not sure yet whether it is worth it?

Then a fixed-price review is the better first step: AI review, a clear recommendation in two weeks →

Questions about the build

What does a build cost?

A fixed price for the agreed scope. There are no list prices, because every project is different. You usually get the concrete quote within five working days of the first call.

We already have a pilot. Do you start from scratch?

Usually not. I build on what you have. If it is unclear whether the pilot is viable, an AI review is the better place to start.

Who owns the code and the architecture?

Code, architecture and methodology stay with you. The handover with documentation and eval suite is part of the delivery.

How is our data handled?

By default, there is no training on client data and no logging to external services without your explicit approval. For sensitive use cases, the system runs on your side or in the EU.

Contact

Email me at info@slworks.de or book a free intro call right away.

Book an intro call

Loads the Cal.com booking calendar (USA). This transmits your IP address to Cal.com. More in the privacy policy (German).