• Distributed AI Systems — Multi-provider LLM routing, Model Context Protocol (MCP), and multi-agent graph pipelines.
• Self-Hosted Infrastructure — High-availability virtualization on Proxmox (Ubuntu VMs), container orchestration via Docker, and secure networking.
• Full-Stack Observability — Telemetry pipelines spanning infrastructure, runtimes, and LLMs using Prometheus, Grafana, Loki, Tempo, OpenTelemetry, and Langfuse.
• Infrastructure as Code (IaC) — Declarative systems provisioning and configuration management via Terraform, Ansible, and automated CI/CD engine loops.

A small editor extension for writing requirements in EARS — the Easy Approach to Requirements Syntax, a convention that keeps each requirement to one of five plain-English templates so it stays clear and testable. EARS is a common format for spec-driven development, including writing the specs that brief LLMs and AI coding tools.

Available on the VS Code Marketplace and Open VSX (so it also installs in Cursor and VSCodium). It works on any .ears.md file and:

• Colours the EARS keywords, <placeholders>, and [DRAFT]/[STABLE] markers as you write.
• Sorts each requirement into its EARS type — Ubiquitous, State-Driven, Event-Driven, Option-Driven, Unwanted — in a sidebar list, worked out from the opening keyword.
• Adds a command to re-organise a spec into tidy per-type sections, a scaffolder for new spec files, and Tab-completion snippets for each template.

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

• File-Driven Pipeline: Decentralized task queue (_queue → _active → _completed) managed entirely through markdown files.
• Unified Engine Matrix: Multi-provider routing layer orchestrating local Ollama nodes alongside Groq, Gemini, Anthropic, and HuggingFace endpoints via a single configuration table.
• Knowledge Synthesis: Complex research execution powered by LangGraph — features parallel web scraping, entity extraction into a KuzuDB graph database, and automated report compilation.
• Self-Healing Workflows: Closed-loop GitLab CI code-generation engine that catches execution context, streams tracebacks back to the LLM, fixes failures, and automatically re-tests.
• State & Memory: Contextual execution memory using a MemPalace MCP server to surface past run data with a single declarative YAML flag.
• Dual-Layer Telemetry: System metrics and runtime traces routed to OpenTelemetry/Tempo/Grafana, paired with deep LLM call, token, and cost analysis via Langfuse. Automated alert routing via Discord.
• Reactive UI: Lightweight administrative interface built with FastAPI, HTMX, and Server-Sent Events (SSE) for live-streaming job states.
• Quality Gates: 87 automated unit/integration tests managed via parallelized GitLab CI and GitHub Actions pipelines with automated deployment on green merge.

Operational risk and structural compliance architecture mapping AI system constraints to automated software guardrails.

• Translates high-level organizational policy directives, regulatory compliance rules, and algorithmic data requirements into auditable code controls.
• Codifies risk-mitigation vectors directly into target deployment configurations and LLM orchestration schemas.

Declarative Prometheus and Grafana stack engineered for multi-node bare metal and virtualized infrastructure visibility.

• Aggregates system metrics, OS logs, and runtime traces across the homelab infrastructure cluster.
• Implements deep per-container resource tracking, performance profiling, and error anomaly detection using standard Docker metrics exporters.

A modular blueprint repo showcasing cloud infrastructure patterns, secure network topography, and automated software delivery.

• Implements immutable infrastructure practices utilizing highly reusable, modular Terraform modules.
• Provisions structured cloud environments alongside automated, repeatable Python deployment pipelines and security baselines.

Enterprise-scoped AI governance framework designed to enforce security constraints, policy alignment, and compliance tracking in automated system workflows.

• My First Alert Fired and It Was Not a Test — 18 Apr 2026
• Git - Green, Blue and Purple? — 18 Apr 2026
• Linguistic Fluidity: Why reading Cyberpunk Makes Me Better at AI — 2 Apr 2026

More articles 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