I like building AI systems that remain useful beyond the demo: grounded, testable, and observable.
I'm completing an M.Sc. in IT & Cognition at the University of Copenhagen. My recent work focuses on RAG reliability, agent workflows, evaluation, guardrails, and human-centred AI. I'm looking for full-time Applied AI or AI Solutions roles in Denmark from September 2026.
| Project | What I built | Evidence |
|---|---|---|
| RAGOps Lens | A production-style RAG evaluation platform with FastAPI, PostgreSQL, pgvector, Qdrant, Docker, and Azure observability. | A 70-case evaluation suite; the calibrated composite gate caught all 5 no-answer cases with a 1.5% false-fallback rate. |
| GitHub Opportunity Miner | A full-stack LangGraph agent that turns GitHub issue evidence into traceable product opportunities and validation plans. | A verified live run processed 38 real GitHub items into one validated opportunity card with no pipeline errors. |
| GazeRAG | A research framework using radiologists' gaze as interpretable anatomical priors for hybrid retrieval. | Gaze-guided retrieval produced small but stable ranking improvements across evaluation settings. |
- Primary: Python — FastAPI, ML/LLM pipelines, retrieval, evaluation, testing
- Working proficiency: SQL/PostgreSQL and TypeScript/JavaScript
- Systems: Docker, Azure, CI/CD, MLflow, Application Insights
I care most about the parts that make AI systems dependable: evaluation, failure modes, evidence, and clear operational signals.