I design and build complete systems — from endpoint and networking to enterprise infrastructure, and now autonomous AI architectures. 18+ years of experience spanning IBM/AstraZeneca-scale consulting, full IT ownership as a CTO, and hands-on systems administration across the legal, healthcare and research sectors.

📍 Sweden · 🌐 Open to remote · 💬 Swedish / English

• Architecture & Infrastructure — end-to-end systems · networking · Active Directory / M365 · Linux / Apache · performance & hardware tuning
• Applied AI — local LLMs · multi-model orchestration · persistent-memory architectures
• Engineering — PowerShell · Python · C/C++ · automation · Windows internals

(https://autom8edIT@github.com/autom8edIT/OmniContext) — ** ("Persistent SRE Data Lake—a massive, specialized MongoDB graph that ingests scattered ops data (logs, registry configs, service states, PDFs) and strips out the noise so any LLM can query a clean, gigabyte-scale reality of the system without hallucinating")

A self-built distributed AI architecture: one persistent, shared "mind" that any model or tool can plug into — from a CLI agent to a local llama-server model — so models build on each other's reasoning and the system compounds over time. If you already use any graph-based memory (Neo4j/Aura), or self-directed model tiering like Constellation it's basically just plug n play, aka. closed-loop debugging.

I started building this because in genereal the standard RAG sucks for SRE. When you ask an LLM to analyze Windows services, you can't just feed it an unsorted 27,000-character text dump. OmniContext ingests any format, structures it into nodes and edges in MongoDB, and gives the LLM exactly what it needs to execute safe operational surgery.

• Public fix accepted to Microsoft's intelligent-terminal (#328)
• Published Chrome extension (TOTP / MFA)
• Contributor to llama.cpp-based tooling

• 📄 CV / Resume: autom8edIT.github.io
• 💼 LinkedIn: linkedin.com/in/joel-larsson
• 📧 Email: joel.larsson@autom8ed.me