Joel Larsson

The end goal is a system that let any IDE, cli anad AI model or agent not only just plugin but benefit from what everyone before has learned.

View on GitHub

Joel Larsson

Solutions Architect · Infrastructure & Applied AI

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


🛠️ What I work with


OmniContext“Persistent SRE Data Lake—a massive, specialized MongoDB graph that ingests scattered ops data 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 general 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.


🏆 Recognition


📫 Get in touch