AI Systems Master
Personal roadmap and reference library for becoming an AI Systems Engineer for regulated European industries by September 2027.
"I build production AI systems for European regulated industries on sovereign infrastructure. AI-Act-grade audit lineage. Sub-3s p95 latency. ~€0.04 per resolved query. Self-hosted models with signed containers. If you need an AI system a regulator can read, that's what I do."
What's here
| Section | Content |
|---|---|
| Playbook | Master career roadmap — 10 pillars, market demand, monetization path, interview readiness |
| Disciplines | 10 deep-reference files, one per skill domain |
| Projects | 3 flagship project specs — implementation blueprints, corpus targets, eval methodology, six benchmark numbers |
The three flagship projects
| Project | Domain | Differentiator |
|---|---|---|
| Enterprise Knowledge Platform | Any / cloud | 10k+ docs, hybrid RAG, eval CI, observability |
| Workflow Automation Platform | Business workflows | Agents, MCP, human-in-the-loop, reliability engineering |
| Insurance Compliance Copilot | Regulated / sovereign | AI Act, PII pipeline, self-hosted models, audit lineage, Cosign |
Project 3 runs on minicloud — the andrelair-platform sovereign Kubernetes stack.
The six benchmark numbers
Every flagship publishes these, with methodology:
- Groundedness rate (%)
- Citation accuracy (%)
- Refusal rate on out-of-scope (%)
- p50 / p95 / p99 latency (ms)
- Cost per resolved query (€)
- Hallucination rate on adversarial set (%)
Platform
Built on andrelair-platform — a 5-node bare-metal Kubernetes cluster (k3s + ArgoCD + Harbor + Vault + Authentik + Longhorn) simulating the IS of a French B2B insurer.
Platform docs: andrelair-platform.github.io/minicloud-platform-docs