I build and troubleshoot infrastructure across the full stack: endpoint, identity, network, operating system, automation, and the operational tooling around it.

I have nearly 20 years of professional IT experience across enterprise consulting, legal, healthcare, research, and managed services. My background spans Windows, Linux, macOS, Active Directory, networking, hardware, and production automation. My current focus is closed-loop SRE:

detect -> diagnose -> remediate safely -> verify

Based in Sweden. Open to global remote and contract opportunities.

I have used ChatGPT since its public launch and continuously evaluate newly released open-weight models from Hugging Face alongside GitHub Copilot Max and SuperGrok. I test them against real repositories, infrastructure investigations, tool use, long-context work, and verifier-backed implementation rather than judging them only through isolated chat prompts.

That experience feeds into two related systems: GodBrain is the Windows research lab, while SREBrain is the clean, professional implementation of the portable SRE contract. The goal is to make the operational workflow independent of any particular paid CLI, GUI, model vendor, or inference engine.

My default working environment is the command line: PowerShell and Bash, Git, APIs, compilers, tests, and repeatable scripts. I use graphical tools when the visual state is itself useful evidence, not as a substitute for automation.

A platform- and model-agnostic project built to create the ultimate SRE agent, with "ultimate" defined as a public set of verifiable engineering criteria.

• Implements a shared Go contract for diagnosis, planning, application, rollback, and evidence.
• Ships fixed read-only diagnostic catalogs for Windows and Linux.
• Executes no user- or model-generated shell commands.
• Produces timestamped SHA-256 evidence receipts.
• Keeps all mutation disabled until an action has an allowlist entry, verifier, rollback path, and failure tests.
• Builds and tests on both Windows and Linux.

A Windows-first local automation lab built around a native C++ control kernel. It is where the deeper closed-loop runtime, local model integration, knowledge boundary, and host-specific experiments are developed.

• Runs an explicit discover -> plan -> execute -> verify loop.
• Keeps model output behind policy and authorization boundaries.
• Uses a Go ingestion and retrieval layer with immutable sources and reviewed operational knowledge.
• Exposes bounded diagnostics, host telemetry, audit history, rollback-aware local edits, and allowlisted remediation.
• Treats the language model as a replaceable reasoning component rather than the security boundary.

The architecture case study that preceded GodBrain and SREBrain. It documents the move from a shared-memory prototype to safer systems with explicit trust boundaries, provenance, verification, and controlled execution.

• Led 20-30 technicians during an enterprise Windows migration for AstraZeneca in an IBM-led delivery environment.
• Participated in infrastructure work for a Swedish Tax Agency data-center project.
• Built complete Windows domains, networks, VPNs, Wi-Fi, servers, and endpoint environments for small and medium-sized organizations.
• Owned IT strategy and operations as the sole technical decision-maker for a law firm.
• Filed a detailed Microsoft Intelligent Terminal bug report that the project team acknowledged as a priority defect.

• Portfolio: autom8ed.me
• LinkedIn: linkedin.com/in/joel-larsson
• Email: joel.larsson@autom8ed.me