Automation and AI for existing processes.
I build automation where your processes already run: between email, CRM, backend and the tools your team uses every day. AI agents take over the steps someone has been doing by hand: sorting enquiries, moving data, drafting replies and offers. They are wired into a pipeline with a human in the loop: at the sensitive points it stops and waits for your approval before anything reaches a customer or a system. What proves routinely right can graduate to running on its own. Scoped and billed per project against an effort estimate, no packages, no flat rates.
What I build.
Workflows across your existing systems
Automated workflows between the tools you already use: email, CRM, accounting, your own APIs. Built with tools like n8n, self-hosted or in the cloud, matching your infrastructure.
n8n consulting and developmentAI agents & pipelines
Agents that classify, extract and draft, wired into multi-stage pipelines. Human in the loop: sensitive steps wait for your approval, routine flows through.
Backend wiring
APIs, data models and background jobs that carry the workflow: Node.js, NestJS, databases.
Agentic development
My own development runs agentically: AI agents write alongside me, review and tests stay with me. From that daily practice I know where agents hold up — and where they do not.
Typical workflows.
Four examples of what these automations look like. Every workflow has a clear input, a clear output and defined points where a human signs off.
Classify and route incoming mail
Emails and attachments are sorted by type and urgency, forwarded to the right people and filed as a case in the right system. Anything the model cannot classify with confidence goes to a human for review.
Structure incoming invoices and delivery notes
Invoices and delivery notes arrive as PDFs or scans. The relevant data is extracted, validated and handed over to accounting in a structured form instead of being typed in by hand.
Prepare customer inquiries
Incoming inquiries are classified and enriched with existing data from your CRM or order history. A draft reply is ready and only goes out after sign-off.
Capture order documents
Orders arrive as PDFs, email text or scans in free form. Line items and terms are recognized and created in your inventory or ERP system.
A different workflow?
These examples show the principle. If a recurring workflow is eating time in your company, we will look at it together. I will tell you in the intro call what can be automated, and what is better left manual.
Book an intro callWhat changes with the size of the company.
The technology is the same in both cases. What differs are the requirements for operations, data handling and approvals, and those shape the project.
Small teams and sole proprietors
- Operations afterwards
- You or one person on the team. The workflows are built to run without me and to be adjusted with a guide.
- Data and models
- Managed n8n or an EU cloud. Language model via API under a data processing agreement.
- Connected systems
- Email, calendar, spreadsheets, a cloud accounting or CRM tool.
- Approvals in the process
- One person reviews what the model cannot classify with confidence.
- Typical starting point
- The one workflow that eats the most time, built in days.
Mid-sized companies with their own IT
- Operations afterwards
- Your IT runs the instance. Handover with documentation, access concept and monitoring.
- Data and models
- Self-hosted on your infrastructure. Which data a model sees is agreed up front with IT and data protection.
- Connected systems
- CRM, ERP, document management, internal APIs and databases.
- Approvals in the process
- Role-based approvals. Traceable who approved what, and when.
- Typical starting point
- Process mapping with the departments, then a pilot with one workflow.
CognitEase
CognitEase is a practice assistant for psychologists and therapists as a Telegram mini app, built and operated by me alone: client management, scheduling and payments — and an LLM pipeline that turns session recordings into structured documentation. Each stage is reviewed before the next one starts, instead of trusting everything to a single prompt.
For client projects, what matters about it is this: I have not just designed an LLM pipeline but built it, put it into operation and keep it stable — architecture, backend, frontend and operations, all single-handedly.
- Role
- Founder
- Stack
- NestJS · Next.js · LLM · Telegram Mini App
- Status
- public beta
- Where does our data run?
- Wherever you decide. n8n runs on your own server or in an EU cloud, and which data a language model sees is agreed before the project starts — with EU hosting or anonymisation where the GDPR or your contracts require it.
- Which processes are worth automating?
- Recurring sequences with a clear input and output: sorting and answering enquiries, moving data between systems, preparing documents and offers. What happens rarely, or differently every time, is better left manual — and I say so in the intro call.
- What does an automation project cost?
- Scoped and billed per project against an effort estimate, which you receive in writing before commissioning. A single workflow takes days; a pipeline with several agents and approval steps takes weeks.
- Do we have to introduce new systems?
- No. The automation connects to what you already use: email, CRM, accounting, your own APIs. Nothing gets replaced — it gets connected.
- Cloud AI or a local model: which fits us?
- The decision rests on four points: how sensitive the data is, how much volume there is, what quality the task demands from the model, and who runs the hardware. For most workflows in a mid-sized company a cloud model with EU hosting, a data processing agreement and anonymisation of the sensitive fields is enough. A local model pays off when data must not leave the building or when the volume is so high that API costs exceed running your own hardware. My own pipelines run on cloud models; a local setup I plan together with your IT.In depth on the blog: AI in the cloud or on your own hardware
- Who runs the automation after handover?
- You do. Every workflow is handed over with documentation, versioning and alerts so that your team or your IT can run it without me. Ongoing support and changes billed by effort are available if you want them.
- n8n or Make and Zapier?
- Make and Zapier are quick to set up and sufficient for simple connections between cloud services. n8n can be self-hosted, versioned and extended with your own code. As soon as data has to stay in-house, workflows branch, or a language model with a review step is involved, that difference is what counts. I use n8n because it can be treated like software: with tests, environments and rollback.
Quote after a free first conversation
First conversation: 30 minutes, free of charge, no presentation.
You describe the situation, I tell you whether and how I can help. No slides, no sales pitch.