About
I'm an AI engineer and technical leader with over seven years designing, developing, and operating enterprise data and AI systems — and Diabase is where I turn that into public, working software.
My work spans full-stack AI application development, language-model fine-tuning, retrieval systems, distributed infrastructure, and production engineering. Day job and independent projects share the same thread: systems that actually ship and stay operable.
Through Diabase, I independently explore and build open-weight language models, AI applications, and privacy-oriented infrastructure — turning research and technical ideas into working systems you can click, clone, or run locally.
I'm based in Sweden. The name Diabase comes from a dense Scandinavian rock — foundations. That still fits, but the portfolio is deliberately broader than “ops only”: models, products, and the judgment calls between them.
Portfolio
Public work a recruiter can inspect without an account — architecture, trade-offs, and honest limits included.
Full-stack AI workspace: conversations, model integration, workflows, and region lockdown. Project write-up with decisions and architecture.
Case study ResearchEmbedding + reranker research previews: language adaptation, contrastive training, leakage audit, corrected splits, MTEB honesty.
Case study ModelsGemma 2 9B fine-tune (QLoRA → GGUF/Ollama), LoRA adapter, and the full model catalog on Hugging Face.
Model lineupFocus
Make it straightforward to build with AI in Europe — without giving up privacy, sovereignty, or control of the stack.
I care about systems you can audit, deploy locally, and explain. Swedish-first models, EU-friendly deployment paths, and honest labels when something is still a research preview.
Connect
Happy to talk models, products, architecture, or roles.
