AI consulting grounded in practice: LLM products that real customers use
I don’t just advise on AI – I build with it. As founder and managing director of ganzheit, I’m responsible for a product in which a large language model is the core of the value it creates. In early 2025, the models simply couldn’t do the core of this product well enough – today they can. So when to wait and when to build is not a theoretical decision for me. And in several client projects, I work right where a demo turns into a tool that runs every day.
Where AI initiatives typically get stuck:
The demo is not the product.
Three examples:
01The untested use case
A use case, a chat window, a first answer that impresses. The team is sure: we’re building this.
Three months of development later, it turns out the model gets the hard cases wrong.
Nobody had defined beforehand what good output even looks like.
02The prototype that never goes live
A prototype that works, and an executive team that’s thrilled.
Then someone asks about the cost per request, about data privacy, and about how you’ll notice when quality starts to slip.
Between demo and production, there are no evals, no monitoring, and no math that adds up.
03AI stays in the chat window
An AI assistant for everyone on the team, and everyone feels more productive.
Still, everyone prompts on their own, while reporting, campaign management, and data analysis run the way they did two years ago.
You’ve gotten faster – but the product hasn’t gotten better yet.
Sound familiar?
Completely normal. Building with LLMs is a different craft from classic software development – and right now, every organization is learning it at the same time.
I bring you the hands-on experience you would otherwise have to gather yourselves.
“After eight years of working together, directly and indirectly, I can say this: in my career, I have met only a few product strategists who think as critically, work as consistently hypothesis-driven, and validate ideas instead of debating them – combined with an exceptionally high bar for product quality, absolute reliability, and a collaboration that is simply fun.”
Learned in production, not in the demo
LLM products that go live
My playbook for LLM applications has six stages, from understanding to scaling, and grew out of ganzheit. Stage 1 shows how you validate an AI use case before the first line of code is written. Stage 2 shows what an MVP really needs: prompt management, tracing, evals, and cost monitoring. The overview shows the whole path. In April 2026, I presented it at ProductTank Hamburg, and you’re welcome to book me for workshops tailored to your team.
AI in your workflows
Becoming more productive is easy; getting better results is not. That’s why, in client projects, I use AI agents where they take over entire steps of the work: I run Google Ads via Claude Code and build LLM-supported data warehousing on Databricks and DuckDB. What this looks like for websites and content is described in two field reports: a static website instead of a site builder and content management without a CMS.
Knowledge that stays with your team
What matters to me is that afterward, you can do on your own what we built together. I work with your people instead of next to them – good AI consulting doesn’t leave a black box behind.
Anyone who claims today to have a textbook for working with AI isn’t telling the truth. The field is changing too fast, and too much is in flux for a cookie-cutter approach.
That’s why I work with you through co-creation: I bring my experience, and together we find solutions that really hold up for you.
Need more than consulting?We also do end-to-end development.
There is a lot of engineering between use case and production. Together with Philipp Kaiser of Build Better, I also build LLM applications myself – the way we do every day at ganzheit, our own startup: from infrastructure to evals to cost controls. Philipp comes from engineering, I come from product. And afterward, your team carries on by itself.
- Product discovery and product delivery
- Pull requests from the whole company
- DevOps
- LLM-based SaaS
- RAG & vector storage
- AI chat with streaming
- Structured outputs
- Evals
- AI Alignment
- Observability with Langfuse
- Profitable AI SaaS business cases
- AI runs Google Ads
- Claim checking
- AI governance in the enterprise
- Enabling teams
Philipp Kaiser, Build Betterbuild-better.de →LinkedIn →
Contact
Want to work together?
Want to know whether AI can already solve your problem today? Then let’s talk about exactly that in a no-obligation intro call – honestly, even if the answer is “not yet.”
Feel free to book a call right away, email me at jan@produktkraft.com, or reach out on LinkedIn.
