A hands-on approach, now applied to software.
{
"name": "Alexander Nilsson",
"location": "Umeå, Sweden",
"role": "System Development Student",
"background": "19+ years, construction industry",
"building": ["Latent Suite", "Lo-Fi AI", "Drop-in Brain", "Mr. Wiggy"],
"focus": "Clean, efficient, privacy-conscious software",
"status": "Open to opportunities"
}Most of what I build defaults to running on your machine. Cloud AI is opt-in, never required.
A strong work ethic carried straight into careful, deliberate engineering.
Clean, maintainable code over quick hacks.
Languages, frameworks, and tools I build with.
A mix of desktop apps and developer tooling I've built and use myself: the Latent Suite for local AI image generation, an AI desktop companion, and a kit that gives AI coding assistants the context to work well.
Search, tag, and de-duplicate your AI generations without touching the cloud. Full-text SQLite search, ComfyUI graph parsing, visual duplicate detection, and offline AI auto-tagging, all running locally.
Stop hand-sorting hundreds of checkpoints and LoRAs. Automatically sorts your model library by architecture using zero-copy .safetensors header parsing, and fetches Civitai metadata for everything it finds.
Turn a folder of raw images into a training-ready dataset. AI watermark removal (Florence-2 + LaMa), multi-format export, and image captioning (Qwen2-VL), all on your own GPU. No uploads, no API keys, no content filters.
The fast, no-frills companion for pulling generation metadata out of locally generated AI images. Recursive parsing across multiple platforms, JSON persistence, and the Latent dark theme, in a JavaFX app that does one thing well.
An always-on-top AI desktop companion that watches your active window and clipboard, proactively offers contextual tips, and answers questions using your choice of Claude, Gemini, OpenAI, or a fully local model via Ollama/LM Studio. Spring Boot REST API backend, Electron + TypeScript frontend.
A portable setup kit that gives AI coding assistants (Claude Code, Copilot, Codex, Gemini CLI) a rulebook and curated per-stack skills before they touch your code, plus a built-in "doctor" that verifies the wiring instead of failing silently. Built on the open AGENTS.md standard, focused on Java/Spring Boot, SQL, and Vue/Angular/Thymeleaf frontends.
A local-first agent runtime with a glass-box dashboard: connects to your local LLM (Ollama, LM Studio) and orchestrates a mesh of small coding micro-agents, planner, coder, debugger, directly on your codebase. Every prompt, diff, and tool call is inspectable in real time, with per-step Git rollback if a patch chain breaks the build.
Reusable, config-driven scraper for news articles and comment sections, with Markdown/PDF report export. Sites, search terms, and comment selectors are managed through an admin UI, not hardcoded, so the same codebase serves different research projects. Includes an AI-assisted site-drafting console backed by a local Ollama instance, never a cloud LLM API.
From the job site to the codebase.
NCC Sverige AB, Umeå
Two decades as a construction worker and tower crane operator: planning, precision, and problem-solving under pressure on major projects around Umeå.
Umeå
Retrained into software development via Higher Vocational Education, focusing on Java, clean architecture, and full-stack engineering.
Open Source
Working on pet projects like the Latent Suite: four local-first tools for the AI image generation workflow, alongside my studies.