I’m Nicholai. I turn emerging AI capabilities into useful systems, with an eye for evidence, reliability, and the people who use them.
A few projects that show how I explore the hard parts: grounding, measurement, integration, and getting software into the world.
An experiment workbench for comparing inference cost, latency, retrieval depth, cache behavior, and answer quality. Built to make model choices more explicit.
A full-stack agent combining live World Bank indicators, PDF retrieval, and LLM analysis, with inspectable evidence and exportable briefings.
A Rasa prototype that checks simulated order data before giving support guidance, with deliberate fallback and human handoff paths.
A small MCP server exposing semiconductor context through named tools and resources, with a rule-based baseline and paired LLM evaluation.
A live FastAPI service on AWS EC2, with Docker, Nginx, HTTPS, GitHub Actions, and a separate Terraform networking lab.
An end-to-end review pipeline combining a fine-tuned DeBERTa sentiment model, topic clustering, and BART summaries in a Gradio interface.
The technology changes quickly. The engineering habits that make it useful matter even more.
Understand the workflow and its constraints before choosing a model, agent, or architecture.
Use baselines, evaluation sets, and clear limitations to learn what actually works.
Connect models to data, APIs, interfaces, deployment, and the people relying on them.
My path into AI came through years of leading teams and improving real operations. That experience shaped how I build: practical first, thoughtful about the user, and comfortable learning in public.
Today I work across LLM applications, retrieval, agents, and cloud delivery. I enjoy the gap between a promising prototype and a system people can depend on, and I’m looking for a team where I can keep growing while contributing from day one.
Conversational AI & retail automation · Rasa project collaboration
Intensive training in machine learning, NLP, LLM applications, and deployment
Team leadership and service operations in The Hague and London
Exploring AI engineering roles, thoughtful collaborations, and interesting problems. I’d be glad to hear what you’re building.