AI Automation & Consulting - from a Developer Who Builds It Himself
No slide-deck consulting: I analyze your processes, build the pilot myself and hand over a workflow that runs - with LLMs, agents and clean integration into your systems. GDPR-compliant, fully EU-hosted on request.
What can actually be automated today
Not every task belongs with an AI. These four areas work reliably in practice.
Support & Communication
Triage incoming requests, generate draft replies, tap into knowledge bases - the human decides, the AI prepares.
Documents & Back Office
Read, check and post invoices, delivery notes and forms into your systems - structured data instead of retyping.
AI Agents & Integrations
Agents that perform real actions in your tools via MCP and APIs - from shop backend to database, with clear guardrails.
Development Workflows
Anchor agentic development in your team: context files, review gates and tooling so AI coding assistants are productive instead of dangerous.
Analysis, pilot, operations - in that order
No year-long project on spec: the pilot shows within a few weeks whether the automation pays off.
Process check
We walk through your workflows and find the spots where automation measurably saves time or errors - and the ones where it doesn't.
Pilot in 2-4 weeks
One tightly scoped workflow goes live: real process, real data, measurable result. Then you decide about scaling up.
Operations & handover
Monitoring, cost control and documentation - and a handover that lets your team evolve the workflow on their own.
GDPR is an architecture question, not a disclaimer
Which data may go to which model? What runs in the EU, what on-premise, what not at all? I plan data flows, data processing agreements and model choice from the start - instead of tacking a disclaimer on at the end.
The consultant is the developer
I don't sell AI strategy slides. What I recommend, I've built myself - publicly verifiable.
My own AI tools on npm
vex (MCP server for the Vendure Admin API), nit (hand UI fixes to coding agents) and wmux (terminal multiplexer for AI agents) are publicly installable.
View vex on npmAI in production at fainin
AI-generated product descriptions run in production on the marketplace - built as a Vendure plugin, not a demo.
Host of the "Vendure + AI" roundtable
I moderate the Vendure community's exchange on AI workflows - over 90% of participants already use AI; the questions are about the how.
Visit the Vendure communityAgentic workflows, documented
vendure-nx ships context files for coding agents out of the box - the same approach I establish in client projects.
vendure-nx on GitHubFrequently asked questions about AI automation
The pilot is deliberately cut small - a low five-figure amount, often less. After that you know the measurable benefit and decide about scaling based on numbers, instead of signing off on a big project upfront.
Recurring tasks involving text, documents or structured data: support triage, document capture, product data, reports. Poor candidates: rare one-off cases and high-liability decisions.
Usually no. Current LLMs with good context (RAG, structured prompts, tool integration) solve most back-office cases without custom training - faster, cheaper and easier to maintain.
Yes. EU hosting, data processing agreements, data minimization and - where needed - local models. Which data an external model may see is decided per workflow, not across the board.
Which process eats the most of your time?
Describe it to me in two sentences. In the process check I'll tell you whether AI helps there - or whether, honestly, simple automation without AI is enough.