Consultancy

Most AI projects don't fail in the demo. They fail in the boring system downstream of the demo. That's the gap I close.

The demo gets the meeting. The demo is not the problem. The problem is the dozen other systems that need to know the demo exists, what it produced, and what to do when it's wrong. AI consultants love demos. The work is everything underneath.

I'm Mischa. I build vertical AI products at Ryde Ventures. The longer story is on about. I take two or three outside engagements per quarter when the work fits and my own products can spare the time.

One process at a time. Transformation programmes are how you spend two years and ship a Confluence page. We pick one process, build the system that runs it instead of your team, and measure the saved time against the fee. If the math doesn't work, I tell you. If it does, we do the next one.

Vertical, not horizontal. A generic AI model knows your industry the way a tourist knows a city. The moat is workflow plus accumulated data plus domain depth. Ten years inside one vertical at Stagent taught me what generic models miss. The work I'd do for you is the same shape as the work I've been doing for myself.

Your data stays where it is. On your servers, in EU hosting, or inside your own cloud. Not in someone's vector database. Not training someone else's foundation model. For procurement and legal, that's a dealbreaker argument. For you, it's the only sane position.

Operator, not slide-maker. The code I'd ship for you is the code I already ship for myself, in Stagent, TAP and Steddle. There is no methodology deck.

Built for agents, not just for people. Stagent answers an AI assistant through an MCP server with 40 tools, where writes are off until a customer turns them on and one test file tries to steal every other tenant's data. Bron serves Dutch law to agents with the source document and a checksum behind every quote. That's the layer most AI projects skip: the part that decides what a model may see, and proves it.

What an engagement looks like

For ops-heavy companies: a one-week audit on site. I embed, map the workflows, interview the people running them, and leave with a prioritized list of the three or five processes where AI saves the most time. Then we build them, one at a time. The standard shape is three processes over five or six months. Fees typically land between €40k for a single process and €300k for a full multi-process programme.

For founders and operators: shorter and lighter. Either a strategy day on your roadmap and AI tooling, or a two-week install where I sit with your team and put the workflow in place you should have had six months ago. Claude Code, an MCP server for the parts of your product that need it, a Ralph loop for the work that wants to run overnight. Fees typically range from €2.5k for the strategy day to €15k for the install.

Exact figures typically get clear after the third coffee.

What I'm not

I'm not a generalist AI consultancy. I don't write 60-page strategy documents. I don't have a partner track or a slide team. If that's what you need, you already know where to find it.

How to start

Email me at mischa@sigtermans.me. One paragraph on what you're trying to do and what you've tried already. If it's a fit, I'll send back a call window inside the week. If it isn't, I'll tell you that and point you at someone who'd be better.

Demos are easy. The systems downstream of them are the job.