I advise companies on adopting AI and build custom AI solutions — from the first idea to a running system. Building software since 2015, many of those years in SAP and ERP environments.
For leadership, IT and business teams who want to use AI where it makes sense: as a custom LLM application, as an AI agent or connected to your systems — with clear access rights and measurable quality.
Services
Analysis, solution, integration, workshop. You choose where to start.
From the first decision to a running system.
01
2 weeks
AI opportunity assessment & advice
We examine your workflows and data and decide where AI pays off — and where it doesn't, including which models fit: cloud, EU-hosted or local. Also for existing AI prototypes that are not yet reliable.
The result Prioritized use cases, an architecture recommendation and an actionable roadmap.
A tailored LLM application or AI agent for one concrete workflow — with your data, an evaluation set and clear success criteria. From pilot to production.
The result A working, tested and documented solution your team can operate.
Hands-on with your own use case: prompting, assistants such as ChatGPT or Claude, your own knowledge bases — and for development teams, coding agents such as Claude Code.
The result A team that uses AI safely and deliberately, with shared rules and next steps.
All prices net, plus VAT. Need an extra pair of hands in a running project? I also join existing teams — remote or in Berlin, around 20–30 hours per week.
The person behind the work
I'm Fabian.
Engineer, consultant and your direct point of contact.
I have been building enterprise software since 2015 — at HARTING, Siemens, Enpal and SAP, for many years with ABAP, RAP, CDS, UI5/Fiori, CAP and BTP. I have worked hands-on with generative AI since before ChatGPT. Today I build my own AI products and work with coordinated AI agents every day.
In my own experiments I study where AI systems go wrong despite good instructions — and how to secure them technically instead of hoping the prompt holds. You work directly with me: from understanding the problem through implementation to handover.
A good fit for: mid-sized and larger companies — especially with an SAP or ERP landscape — that want to adopt AI, build a concrete custom solution or make an existing prototype production-ready.
Professional background
Experience with systems. And their everyday use.
2024–2025
SAP
Product Compliance development
A core SAP module with long-lived enterprise processes and high quality requirements.
My contribution: Built backend features with ABAP, RAP and CDS and user interfaces with UI5/Fiori, explored GenAI use cases and led workshops within the team.
2023
Enpal
Full-stack development
A solar customer portal: helping customers navigate a complex product and service journey.
My contribution: Enhanced the customer portal through full-stack development, making information easier to understand and act on.
2022–2023
SAP
Prototypes & generative AI (working student)
A dynamic environment for new product ideas.
My contribution: Built prototypes with Python, CAP, BTP, Java and Node. Main use case: extracting the EU Taxonomy criteria with language models and RAG into a structured data catalogue. Also introduced a fellow working student to generative AI.
2020–2022
Siemens
Mobile apps & dashboards
Operational data for the team's day-to-day decisions.
My contribution: Developed mobile applications and operational dashboards that made live data usable in everyday work.
2015–2018
HARTING
SAP development
Custom SAP development for business processes across the web shop and operations.
My contribution: Built custom SAP solutions with ABAP and UI5/Fiori, connected SAP ERP with web applications and data tools, and onboarded apprentices into the team.
Company names describe experience, not client endorsements.
A command-line AI agent that binds every approval to the exact reviewed result and stops superseded runs. In a pre-registered experiment: 0 of 30 rule violations, against 8 of 29 for an identically instructed baseline agent. A research instrument, not a finished product.
Generative GPU visuals for DJ sets, played live from the keyboard. An autopilot follows the beat, and the stage window keeps running even if the controls crash. Runs entirely in the browser.
A music studio in the browser: arrangement, instruments and mix can be edited visually or as code, both on the same project. Its own audio engine, with WAV, MIDI and stem export. In development.
Audio engineTypeScript
How we work together
Start small. Keep improving.
01
Understand the work
We start with your people, the workflow and its obstacles. Together we define how you will recognize an improvement.
02
Test the smallest useful solution
We pick one use case, check data, models and risks, build a pilot and compare it with how you work today.
03
Build for everyday use
What holds up becomes production-ready: with tests and evaluation, clear access rights and human oversight in the right places.
04
Enable your team
I train the people who work with it, document decisions and hand over ownership. We agree how you will measure and improve the solution.
Before we start
What you should know.
How do you protect our data? Where does it live?
I work in your environment with the minimum access needed. By default I use EU-hosted models, such as Azure OpenAI in EU regions or Mistral. Your data is not used to train models. For personal data we conclude a data processing agreement (DPA), and I'm happy to sign an NDA as well.
Can we use AI without the cloud?
Often, yes. In the assessment I compare cloud, EU hosting and locally run open models by quality, cost and data protection, and set up a pilot where useful. Long-term operation of your own GPU infrastructure is planned together with your IT or a specialized partner.
Do you train your own AI models?
No. I use proven language models — through provider APIs, EU hosting or locally — and build the actual solution around them: connection to data and systems, prompting, agent logic, evaluation and operation. Vendor-neutral and matched to your existing systems.
What does the EU AI Act mean for our project?
Most projects fall into the minimal- or limited-risk categories. In the assessment I classify the use case and document transparency obligations. Since February 2025 the AI Act also requires sufficient AI literacy among staff — the workshop helps with that. This is not legal advice.
How long does a project take, and what does it cost?
The AI opportunity assessment takes 2 weeks and starts at €4,900 net. Custom solutions and integrations are offered at a fixed price after a short scoping. The team workshop takes 1 day and starts at €1,900 net. Work inside your team is billed by effort.
A responsible person, access to sample data or documents (anonymized is fine) and 2–3 hours per week from the people who know the workflow during the assessment. For the go/no-go decision we need someone with decision authority.
Your next step
Which workflow could AI make easier?
Tell me what is slow, repetitive or needlessly complicated — or which AI idea you have wanted to build for a while. You don't need a finished concept.