Selected work

An AI-powered monitoring platform built around a simple idea: describe what you want to track in plain English, pick a data source, and let the system handle the rest. A scheduler dispatches polling agents that fetch new items from external APIs - Swedish job listings, parliamentary documents, EU energy market data - then run an LLM eval against your monitoring prompt and route matches via email, webhook, or the in-app dashboard. Runs locally with Ollama (llama3.1) as the default model, with Claude as an optional cloud fallback - keeping costs at zero by default. A built-in review queue lets you audit and correct AI decisions to improve future runs. The admin panel covers source health, per-run LLM cost tracking, prompt management, and user management.

A fully automated daily pipeline: it pulls from 18 RSS sources, then ranks, categorizes and summarizes the stories with Claude Haiku. The top picks are published to a live feed and delivered as a morning email digest - orchestrated by GitHub Actions and Vercel Cron, with in-house double opt-in over Resend and Supabase.

An AI-assisted job-hunting workspace built on a multi-agent pipeline. One agent scrapes and ranks listings against your profile; a second scores your CV fit with concrete strengths, gaps, and next steps; a third drafts a tailored, language-aware cover letter. Powered by Pydantic AI + Groq with SSE streaming, and it runs entirely locally - everything stored as JSON, no database.

Paste your CV and a job posting URL and get an honest, AI-powered fit analysis in seconds. It parses the posting and your CV (PDF or plain text) and returns matched strengths, missing requirements, and an overall verdict - in English or Swedish.

A retrieval-augmented chat tool for your own documents. It ingests a set of PDFs, builds local embeddings, and answers questions grounded strictly in their content - so replies stay accurate and cite the source material rather than hallucinating.