Joel Larsson

Senior SRE and Infrastructure Automation Engineer

I diagnose and automate infrastructure across endpoint, identity, network, operating system, and application boundaries. I bring nearly 20 years of professional IT experience from enterprise consulting, managed services, legal, healthcare, and research environments.

**Sweden

Open to global remote and contract opportunities

Swedish and English**

View resume | GitHub | LinkedIn | Email

What I deliver

Root-cause analysis across layers

I work from evidence rather than repair checklists: client symptoms, DNS and routing, identity, services, operating-system state, logs, hardware, and the application path. The goal is not merely to restore service, but to identify the failed layer and prove that the repair worked.

Infrastructure automation

I turn repeated operational work into bounded, observable automation using PowerShell, Bash, Python, C/C++, Go, and Rust. I favor explicit inputs, dry-run or preview stages, audit history, rollback, and post-change verification.

Enterprise systems

My background includes Windows and Linux administration, macOS, Active Directory, Microsoft 365, Entra ID, Intune, PKI, VMware, Hyper-V, DNS, DHCP, routing, switching, VLANs, VPN, Wi-Fi, servers, storage, deployment, and endpoint engineering.

Applied AI for operations

I build local-first agent systems that retrieve operational knowledge, inspect a host, propose bounded actions, and verify outcomes. The language model is a replaceable reasoning component. Authorization, policy, execution, and audit stay in deterministic code.

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 compare them on real SRE investigations, repository-scale coding, tool use, context retention, and verifier-backed implementation rather than on chat impressions alone.

CLI-first engineering

My primary interfaces are PowerShell and Bash, supported by Git, APIs, compilers, tests, and repeatable scripts. I use graphical interfaces when they expose a signal that is inherently visual; routine administration and engineering should remain observable, reproducible, and automatable.

Featured project: SREBrain

SREBrain is a platform- and model-agnostic project built to create the ultimate SRE agent. In this project, “ultimate” is not an unqualified superlative; it is a public set of acceptance criteria covering diagnosis, authorization, verification, rollback, evidence, and platform support.

Engineering problem

Traditional support tools inspect isolated components. Generic AI assistants can produce plausible advice but usually lack trustworthy host evidence, controlled execution, and a verifier. SREBrain separates the portable operational contract from any one model, interface, or operating system.

Design

detect
    -> diagnose
    -> plan
    -> authorize
    -> apply
    -> verify
    -> record

Current capabilities

Shared Go interfaces for diagnostics, plans, application, rollback, and evidence.
Fixed read-only Windows and Linux diagnostic catalogs.
No shell construction from operator or model text.
Timestamped SHA-256 receipts for exact diagnostic output.
Mutation methods that fail closed until the complete verifier contract exists.
Windows and Linux CI from the first commit.

The first release is intentionally diagnose-only. A mutating action will ship only when it has a checked-in allowlist entry, typed plan, authenticated policy decision, postcondition verifier, rollback path, and failure fixtures.

Explore SREBrain

Research lab: GodBrain

GodBrain is the Windows-first research lab where the deeper closed-loop runtime, native C++ kernel, local model integration, operational knowledge boundary, and host-specific experiments are developed.

SREBrain extracts the portable professional contract. GodBrain remains free to push Windows B-line automation and local-agent research without making every experiment part of the recruiter-facing product.

Architecture case study: OmniContext

OmniContext preserves the earlier prototype and explains the engineering lessons that led to GodBrain and SREBrain.

The case study covers:

why direct model access to memory and execution was rejected,
why raw data and reviewed operational knowledge need separate trust levels,
why verification is more valuable than adding more agent nodes,
how a model-specific prototype became a model-agnostic kernel boundary,
how privileged automation can remain useful without becoming an uncontrolled remote shell.

Experience highlights

Led 20-30 technicians in an AstraZeneca enterprise Windows migration delivered in an IBM-led environment.
Participated in infrastructure work for a Swedish Tax Agency data-center project.
Built Windows domains, networks, servers, VPNs, and Wi-Fi environments from zero for small and medium-sized organizations.
Owned IT strategy and daily operations as the sole technical decision-maker for a law firm.
Supported demanding legal, healthcare, research, and enterprise environments.
Reported a reproducible Microsoft Intelligent Terminal defect that the product team accepted as a priority issue.

How I work

Establish the observable failure and its blast radius.
Identify the layer that can explain the evidence.
Run the smallest distinguishing probe.
Apply the least invasive reversible correction.
Verify the original symptom and dependent services.
Record the evidence so the next incident starts from known facts.

This is also how I build automation: the verifier is part of the feature, not an afterthought.

Contact

I am interested in senior SRE, infrastructure automation, Windows/Linux platform, and reliability engineering roles that support remote work from Sweden.