Full-stack AI/ML & Data Engineer who builds and ships GenAI applications and production ML systems end-to-end, fast and cost-efficiently, using Claude and coding agents to accelerate delivery. Freelance AI/ML engineer since 2021, delivering for clients including Klarna (5,000+ employees), Booking.com, and ING, and founder of PrimAxiom Labs, an Amsterdam venture building and operating AI-driven software products. Experienced across multi-agent systems, RAG pipelines, LLM fine-tuning and applications, recommendation systems, NLP, and data infrastructure. Led engineering at Vamp.io (acquired by CircleCI) managing a 12-person team. Open-source contributor (llama.cpp sampler, vector search) with expertise in Python, PySpark, LangChain, Hugging Face, scikit-learn, and distributed systems.
Focused open-source samples extracted from production work — clean, tested, CI-backed reference implementations of larger systems.
Local meeting-intelligence agent: Whisper transcription → structured brief (summary, decisions, action items with owners, topics) → Q&A grounded in the transcript. Ships as a library, CLI, and MCP server with pluggable STT/LLM backends and structured (JSON-schema) output.
"A code review, but for UX": drives Playwright and a vision LLM to review a URL from each persona's point of view — scoring the page, listing problems, and tracking scores over time, with a CI merge gate. Pluggable local-first vision backends (Ollama, Qwen2.5-VL); library + CLI.
MCP (Model Context Protocol) server that gives AI agents and assistants (Claude Desktop, Claude Code, Cursor) a multilingual place-extraction tool — cities, countries, regions, and landmarks across 13 languages — backed by the mDeBERTa/ONNX models below.
Open-source Python library and CLI that generates synthetic labeled data for NER / token classification — from templates + gazetteers or local-LLM few-shot with structured (JSON-schema) output — with balancing and span-preserving augmentation. CI-tested against llama.cpp and Ollama; the data-generation approach behind the place-extractor models.
Multilingual place and entity extraction models built on mDeBERTa for token classification across 13 languages, published on Hugging Face with ONNX-optimized versions for fast inference.
Vector search engine built in Go for ML systems. High-throughput feature serving with low latency. 54+ GitHub stars.
Developed custom sampler in llama.cpp enabling strict constrained sampling in LLMs. Library for constrained text generation with Large Language Models for structured output using GGUF models.
Contributor to Spotify's approximate nearest neighbors library for vector search. Built custom implementations for production use cases.
Focus: Machine Learning applied to signal/image/audio processing. Developed classification models using SVMs, including full ML pipeline: data collection, feature engineering, model development, testing, and validation. Technologies: MATLAB, Python, Support Vector Machines. Worked as Research and Teaching Assistant.