Mike Gölz.

For over a decade I have owned data-warehouse and ETL pipelines in production: from initial source analysis to the daily refresh that must not fail. With everything that goes with it: data-quality monitoring, consistency checks across multiple source systems, consolidation logic for several tenants. Power BI and Cognos as endpoints, once the data model is clean. What I learned along the way: eighty percent of reporting problems are not SQL problems, they are data-model problems. Skip that layer and every new KPI becomes a new exception.

SQL tuning on MySQL, PostgreSQL, and DB2 is a daily tool, not academic knowledge. Window functions instead of self-joins, common table expressions for readability, execution-plan analysis before any index decision. Materialized views and partitioning for large aggregates. Reproducible pipelines with a validation layer so a broken source fails loudly instead of silently producing garbage. Audit trail on every refresh as standard, not an afterthought.

Refactoring grown Excel and Access VBA estates from fifteen to twenty years of operational use. Concretely: Option Explicit everywhere, typed ranges, custom error classes with logging, exported modules under version control. Backend migration of Access front ends to SQL Server via linked tables when performance or multi-user contention demands it. Power Query as a reproducible ETL layer with query folding to source, parameter tables, custom function library, schema validation via Try/Otherwise. VBA stays only where Excel is genuinely the best tool.

For local AI, I run setups with Ollama, llama.cpp, and vLLM for data that cannot leave the building. Concretely: GPU sizing, quantization choices, embedding selection like bge-m3 or multilingual-e5, token budget planning. RAG architectures with Qdrant or pgvector, n8n as on-prem orchestrator, MCP servers as tool bridge to existing systems. GDPR, processing agreements with model providers, EU AI Act risk classification, and DPIA are part of the architecture, not bolted-on compliance cosmetics.

Operational ITSM experience from real organizations, not from slides: KPI reporting on existing data models, IAM architecture, Citrix rollouts in the thousands of users, audit-capable access concepts. 1st and 2nd level support as part of the same stack, not a separate world. What I learned: every system is only as good as the handover documentation that carries it. Without that, the system runs on trust, and trust is not an architecture principle.

Handovers are part of the delivery with me, not an afterthought. Concretely: README with setup, assumptions, known limits, maintenance conventions. Code walkthrough with recording. Version control including all exported VBA modules or M queries. The goal is always the same: a second developer onboards in three to five days, no key-person risk, no black box.

Multi-year independent trainer and workshop business with a complete service cycle: needs analysis, design, delivery, follow-up. Excel, VBA, SQL, Power Query, reporting, and AI blocks for mixed audiences from business teams to IT. In-house, remote, or hybrid, depending on the team and the data situation. Training is not a one-way lecture but work on the participants' real systems.

The background is the application developer apprenticeship with a production web-database project as the final piece, full-stack responsibility from data model to UI. Same discipline since then for every engagement: understand first, build second, document third. In that order, no exceptions.