Data warehouses and the AI products that run on them, from source system to interface. One pair of hands, and it runs in production. See fig. 01.
Part 02The dimensional model: facts, dimensions, history, lineage. Most of the time goes in here and nobody talks about it. Hence the thickness of this plate.
XBAS Business Intelligence · Almere NL · Doc 2026-07
Most projects start halfway down this list and get stuck on what is missing below it. I build all three, usually from the bottom up: part 02 in fig. 01 carries the rest.
DWH
Data warehouses, pipelines, models. The foundation that reporting and AI have to stand on. Without this part, the rest is a demo.
AI
Language-model applications that survive production: grounded in your own data, evaluated on answer quality, deployed in your own environment.
APP
The shell around it: the interface where the work becomes visible. Deployed quickly, and built so someone else can pick it up later.
At the top, what runs in production at clients. Below that my own products, and at the bottom a few demos built in an evening to show what is possible in a short time. That difference is large enough to state outright.
In production
Dutch trade union (name withheld) · 2026
An AI knowledge assistant covering hundreds of collective labour agreements (cao's), in daily use at a Dutch trade union. Every answer comes with a clickable source in the document.
More in production at clients
In production
A Dutch trade union (name withheld) · 2026
A daily sync that puts survey data from an external platform into the data warehouse of a Dutch trade union, with an audit trail of every change per answer.
In production
A Dutch trade union (name withheld) · 2026
An outdated monthly database restore replaced by a daily API feed that updates administrative and budget data in the data warehouse, with full history through temporal tables.
In production
A large Dutch company (SAP S/4HANA) · 2026
An announced SAP security patch blocked the protocol used to pull financial data out of S/4HANA. Instead of falling back on a slower replacement, the extraction was rebuilt on a compliant table connector, with change detection built in-house.
Delivered
A large Dutch company · 2026
An internal BI dashboard that watches Azure Data Factory pipelines and restarts them on request, where each user only sees the tiles that matter for their role.
Delivered
A large Dutch company · 2026
An AI agent that answers questions about the data warehouse in plain language, alongside the existing Power BI. The language model never touches the data directly; it only calls grounded tools behind a security layer.
In production
Flevo Feestverhuur, party and event rental (Amsterdam-Almere area) · 2026
A bilingual rental website with a quotation flow, a deposit through iDEAL and an admin environment of its own in which the owner runs stock, planning and finances themselves.
In development
Lowe's Mobility Care Transportation, Virginia (US) · 2026
An accessible site for an American medical transport company, with an AI-assisted CMS: the owner changes text through a chat assistant that stays within hard limits.
Own products
2026 · In development
A read-only tool that runs against your SQL Server or Azure SQL warehouse and returns three layers: AI-generated documentation with a lineage graph, a deterministic audit of dead code and orphaned references, and a compact AI context file. All output is self-contained HTML, and your data does not leave your own Azure tenant.
2026 · Ongoing, in use on this site
A self-built replacement for Cal.com that pulls free slots out of Google Calendar, books them with a Meet link and lets visitors cancel through a signed link. No database, and with deliberately narrow calendar permissions.
Demos · own initiative
2026 · A single sitting
Trilingual site about a Sierra Leonean politician, with a chatbot that answers only from her own public statements.
2026 · A single evening
Bilingual concept site for an eco-retreat on the coast of Sierra Leone, built around a hand-drawn, interactive master plan.
Hiring a single developer raises a few fair questions. The answers are here, including the ones that count against me.
Point of contact
No account manager between you and the code, no context lost in a handover. What you discuss is what I build.
Your data
Your data stays inside your own Azure or on-premise environment. For DocumentChat that was a hard requirement from the client, and therefore the architecture: no document text leaves it.
Continuity
That I work alone is a real risk, and a promise does not solve it. What does: documentation, a mainstream stack and code the next developer can pick up. You are not locked in.
AI
AI writes alongside me, and that saves weeks. The design, the data models and the trade-offs are mine, and I can account for every one of them.
I have been building data warehouses for over fifteen years. For retail groups, trade unions, property managers and a handful of others who need to know what their figures actually mean.
In recent years the work has shifted. Those same foundations now feed AI applications: an assistant that really knows an organisation's collective labour agreements, a semantic layer that makes the AI and the reporting say the same thing. The web shell brings it together so that people can do something with it.
Stiekema Studio is where those three come together. One person, AI as co-builder, and work that runs in production.
Not a sales call. You tell me what you want to build or what is not working, I tell you what I would do and whether I am the right person for it. If I am not, I say so.
You pick a time from my calendar in the form on this page. It is booked straight away, with a Google Meet link and a link to cancel if something comes up. No back and forth about dates.
Prefer another way
Request · introductory call · 30 min · Google Meet
Fetching the calendar...