Infrastructure · Observability · AI Workflows

regulation → system → constraint → control → automation

I build self-hosted infrastructure, monitoring stacks, and AI tooling from my homelab. Most of what I ship is open source.

📍 Macclesfield, UK
🔗 davidcockson.com · blog · linkedin · hello@davidcockson.com

• self-hosted infrastructure on Proxmox + Docker
• observability (Prometheus, Grafana, Tempo, OpenTelemetry)
• infrastructure-as-code (Terraform, CI/CD pipelines)
• distributed AI systems (LLM routing, MCP, agent workflows)

Self-hosted LLM job runner that turns an Obsidian vault into a distributed AI workbench.

• file-based job queue (_queue → _active → _completed) driven from markdown
• multi-machine model routing across a VPS and home server
• semantic memory via MemPalace MCP — past outputs retrieved with one YAML flag
• OpenTelemetry traces to Tempo/Grafana, Discord alerts on failure
• FastAPI + HTMX dashboard with live SSE output streaming
• 76 tests · GitLab CI · auto-deploy on merge

→ davidcockson-compliance/vault-runner

Monitoring tool for the UK Gambling Commission licence register — domain discovery, corporate group identification, compliance views.

→ repo · live

Converts the UK Gambling Commission's LCCP regulations into a structured dataset with filtering and gap-analysis tooling.

→ repo · live

Prometheus + Grafana stack for a home server. Host metrics and per-container visibility via Docker exporters.

→ repo

Terraform and Python automation practice — IaC patterns, CI/CD pipelines, repeatable cloud deployments.

→ repo

Structured governance models for AI systems, generated and maintained with AI-assisted workflows.

• Sable AI Governance Framework
• Pickles GmbH AI Governance Framework

• Did the Supply Chain Attack Compromise Me? A Morning Panic Check — 31 Mar 2026
• LCCP Filter — Self-Made Tools — 26 Mar 2026
• Monitoring & Observability Journey — 25 Mar 2026

More at blog.davidcockson.com →

flowchart LR
A[Observe system] --> B[Find constraint]
B --> C[Map the gap]
C --> D[Design control]
D --> E[Automate solution]
E --> F[Monitor outcome]
F --> A

The loop applies whether the system is regulatory, operational, infrastructural, or an AI workflow. Systems fail at the constraint. Find it, control it, automate around it.

Eight years inside complex regulatory systems, mostly in the UK gambling industry. The work wasn't policy — it was finding structural weaknesses, investigating systemic failures, and designing operational controls to stabilise things.

The questions were always the same:

Where is the constraint?
What breaks first?
What control stabilises the system?

Those questions work just as well on infrastructure, pipelines, and distributed systems as they did on compliance frameworks.

Regulatory systems analysis
        ↓
Structured regulation datasets
        ↓
Compliance tooling (React + data models)
        ↓
Automation and deployment (Docker / Cloudflare)
        ↓
Cloud infrastructure experimentation
        ↓
Monitoring and observability systems
        ↓
AI infrastructure and agent workflows

Each step builds on the previous one. Direction of travel: analysis → tooling → infrastructure → automation.

• systems fail at the constraint
• the goal isn't complexity — it's stability when nobody is watching
• automate the boring stuff; monitor everything that matters
• ship small, measure, iterate

Map the gap. Architect the control.