DevOps Lead and software engineer. 13+ years building software, the last 5+ running the platforms it ships on.
I build infrastructure that stays calm in production, and open-source tooling for self-hosting and for working with AI coding agents. I write about both at nasraldin.com.
Website · LinkedIn · X · Credly · RSS
Diwan — a software factory for AI coding agents. A system for planning, assigning, supervising and verifying work done by coding agents such as Claude Code, Codex, opencode and Antigravity, from design through to a tested change request. Tasks move across an agentic Scrumban board, a supervisor hands out only the work whose dependencies are ready, and a human approves every merge. The write-up is on my blog.
Local AI on your own hardware. ai-lab is the CLI I use every day to run, profile and monitor local models with Ollama on Apple Silicon.
Self-hosted developer platforms. One-command, reproducible stacks for Git, CI/CD, runners and object storage, published below.
| Project | What it is |
|---|---|
| ai-lab | Local AI development CLI for Ollama on Apple Silicon: model profiles, performance diagnostics, a lightweight coding agent, repo-aware chat, and an optional health-check daemon. |
| github-runner | Manager for self-hosted GitHub Actions runners across multiple repositories and pools, on Linux (Docker), macOS and Windows, x64 and ARM64, with ephemeral runners and automatic re-registration. |
| gitlab-ce | Self-provisioning GitLab CE on Docker Compose: container registry, MinIO object storage, five auto-registered CI/CD runners and a shared S3 cache, with no manual setup steps. |
| OneDev | Production-oriented, headless OneDev stack (Git server, CI/CD, Kanban, packages) with PostgreSQL, a dedicated build agent, and Docker socket access isolated to the runner. |
| MinIO-AIStor | Docker Compose deployment of MinIO AIStor as an S3-compatible store for objects, tables and files, with single-node and four-node distributed topologies. |
| iHosts | macOS desktop app for managing /etc/hosts: inline editing with validation, search and filters, automatic backups, and version history with restore. |
- Agents do the typing; engineers own the outcome. Every agent-written change goes through the same review, tests and pipeline as any other change, and a human approves the merge.
- Plans before prompts. Specs, rules and tasks live in the repository, so any agent or person can pick up the work with the same context.
- Orchestrate, don't babysit. Several agents work in parallel on dependency-ready tasks, with usage and cost monitored.
- Local first where it makes sense. I run open models on my own hardware for private and offline work, and use hosted models where they are the better tool.
| Area | Tools |
|---|---|
| Cloud and platform | Azure (AKS, APIM, Firewall), AWS, Kubernetes, Docker, Helm, Terraform, Nginx, Proxmox |
| Delivery | GitLab CI/CD, GitHub Actions, Azure DevOps, ArgoCD, self-hosted runners |
| Observability | Prometheus, Grafana, Elasticsearch, Kibana, Azure Application Insights |
| Identity and workflow | Keycloak, Camunda 8 |
| AI engineering | Claude Code, Codex, opencode, Antigravity, Ollama, LM Studio, Open WebUI |
| Languages | TypeScript, JavaScript, Python, C#, Dart, Shell |
| Application | Node.js, .NET, React, Next.js, Vue, Angular, Electron, Flutter, GraphQL |
| Data and messaging | PostgreSQL, SQL Server, MySQL, MongoDB, Redis, Cosmos DB, RabbitMQ, MinIO |
Long-form notes on engineering, reliability and calm production at nasraldin.com. Recent posts:
- Building a Software Factory for AI Coding Agents (Codename: Diwan)
- Ollama vs LM Studio: Which One Should You Use?
- Local LLMs for Beginners: Your First Steps
- Linux for Beginners: From Zero to Actually Using It
The fastest way to reach me is the contact page or LinkedIn. Issues and pull requests on any of the projects above are welcome.
Focus on your goals and always move towards them.





