AI-driven ERP for our own manufacturing
FERNAU Präzisionstechnik GmbH runs an ERP with 17 AI agents that I developed myself. Order entry takes 26 seconds instead of twelve minutes, measured in live operation.
Find · test · build
I help you find where AI really saves time in your business, and then I build the solution. I do this with Timo Külbel, a software developer with more than 25 years of experience. Where it makes sense, the AI runs locally at your company, without data in third-party clouds. My own projects show that this works.
My projects
FERNAU Präzisionstechnik GmbH runs an ERP with 17 AI agents that I developed myself. Order entry takes 26 seconds instead of twelve minutes, measured in live operation.
Monitors German federal law and EU law, detects changed provisions and summarises them in plain language. It also manages obligations, risks, documents and evidence to ISO/IEC 27001 and 42001. The AI runs entirely locally, and a human decides on every suggestion.
Dictation software whose speech recognition runs 100 percent locally, with no cloud and no account. Core path signed off on the Mac; for Windows, field testers are still wanted. Open source under the MIT licence, the code is public on GitHub.
About VoiceWall Source code on GitHubAn audit recorder for AI coding assistants. It records every observable action in a tamper-evident way and blocks dangerous commands before they run. Everything stays on your own machine, with no cloud and no account.
Self-hosted meetings with audio and video in the browser. Recording only starts once everyone has agreed, each voice gets its own track, and transcription runs locally on your own server.
A self-hosted AI platform for tax advisers, audit firms, law firms and public authorities. Local is the default. If something does go to an EU cloud in hybrid mode, it only passes through a guard that cleans the data first.
My own bookkeeping, local on the Mac and with local AI: reading receipts, suggesting accounts, preparing e-invoices and advance VAT returns in an auditable way. With an audit trail that cannot be altered.
Each case moves as a folder from enquiry to archive. A fingerprint is stored for every file, and nobody can change or delete the log.
On top of that come tools for our own operations, such as predictive maintenance and a knowledge bot for the shop floor.
How it works
Together we look at where time is lost in the business: in order processing, in administration, at the machine. I come from manufacturing and know such places from my own daily work.
Before much money is spent, a small prototype with your real data shows whether the idea holds up and pays off.
Timo Külbel and I build the solution in manageable steps. Each step is tested before the next one begins.
Training for your team, documentation and handover to live operation.
On request, models and data run on your own machines. What goes to a cloud is agreed together beforehand.
Because I audit AI systems, we build them auditable from the start: with classification under the EU AI Act, logging, access control and tests.
Automated tests are part of every version we ship.
Model and technology follow your task and your existing IT.
How the classification under the EU AI Act works is described under EU AI Act readiness.
Use cases, business logic, classification
Managing director of FERNAU Präzisionstechnik GmbH, ISO/IEC 42001 Senior Lead Auditor and Senior Lead Implementer (PECB). Finds where AI pays off, owns the processes and business logic, and makes sure the solution fits the EU AI Act and ISO/IEC 42001.
About meArchitecture, code, security, tests
Software developer with more than 25 years of experience in full-stack development and IT consulting, founder of Parlison Code Couture UG. Responsible for software architecture, code quality, security and automated tests.
Your contracting party is FERNAU Präzisionstechnik GmbH. Timo Külbel works with us as an external partner.
I do not audit what we have built for a company: no audit and no Audit Ready seal assessment at the same company. That needs an independent party. If a company I audit uses software I built, I say so before the engagement and do not assess that software there. How I separate my roles is set out in the impartiality statement. If you want ongoing support for AI in your business, see the external AI officer.
That depends on the task and the scope. It starts with a free first call. Afterwards you receive a quote with a clear scope and price, for larger projects first for a small prototype that shows whether the idea holds up.
No. Many tasks can be solved with AI models that run on your own machines. My legal monitoring system and VoiceWall work exactly like that. Where a cloud service fits better, we say so openly and agree beforehand which data goes there.
Yes. As a rule, a new solution connects to existing systems, for example the ERP, file storage or email. Even an old ERP without an interface can often be automated instead of replaced. We clarify how that works in your case in the first call.
We clarify that before the start: who takes on maintenance and further development, and what the handover to your IT or a service provider of your choice looks like. Documentation is always part of it.
The classification is part of the project. At the start we clarify whether and which obligations under the EU AI Act apply, and we build in labelling, logging and human oversight where they are required. This does not replace legal advice.
No. Whoever built it does not audit it independently. Afterwards I do not take on any audit or Audit Ready seal assessment at the same company. That needs an independent party.