Portfolio
A selection of projects I’ve worked on.
Company Knowledge Assistants (RAG-based LLM Systems)
For several companies in the Netherlands and the United States I designed and implemented “company LLMs” – internal assistants that understand their documentation, processes and historical data. These tools help teams find answers faster, draft content, and navigate complex internal knowledge bases without having to know exactly where things are stored.
The systems use retrieval-augmented generation: relevant documents are retrieved from a vector index and passed to an LLM, so answers stay grounded in the company’s own data. Depending on the client, the model is either a hosted API (e.g. OpenAI / Anthropic) or a self-hosted open-source model such as Llama 3.
Tech: Python (FastAPI/Flask), hosted and self-hosted LLMs (e.g. Llama 3 via vLLM / text-generation-inference), SentenceTransformers embeddings, vector databases (Qdrant, Pinecone, Weaviate, Chroma), S3-compatible storage, Docker-based deployment, and simple React/Next.js front-ends with company SSO (OAuth/OIDC) for internal access.
Jira Issue Assignment Intelligence – Atlassian
As part of the Jira development team in Amsterdam I worked on core product features and reliability, and helped explore how AI could make teams faster. I built an early prototype of an AI-powered assignee suggestion system that analysed issue history, team structure and workload to recommend the most suitable person for a new ticket.
The work combined Jira’s existing workflows with lightweight machine learning models and careful UX so that suggestions felt helpful rather than intrusive, and always kept humans in control of the final decision.
Tech: Java, Jira platform APIs, internal data services, ML experimentation (Python), collaborative product design with PMs and designers.
Product: Atlassian Jira
ERP and Operations Platforms – Bizboard, Sunvigo, SpotmyEnergy
For several companies in the Netherlands and Germany I’ve helped design and build ERP-style systems that tie together sales, operations, billing and technical data. The goal in each case was the same: fewer manual steps, less copy-pasting and more reliable, traceable processes.
This has included extending existing systems and building new services around them, integrating data from CRMs, energy devices and external partners, and giving teams simple tools that reflect how they actually work.
Tech: Python (including Odoo), TypeScript, PostgreSQL, data pipelines, REST APIs, containerised services (Docker/Kubernetes).
Native Mobile Apps for Dutch Companies
Over the years I have built and maintained native Android and iOS applications for several companies in the Netherlands, ranging from small tools to more complex, data-driven apps. The focus has always been fast, smooth experiences and clean integration with back-end services.
Early work involved Java and Objective-C, later moving to Swift and modern Android tooling. In all cases the apps were closely aligned with the client’s workflows and branding rather than generic templates.
Tech: Java (Android), Objective-C, Swift, REST/JSON APIs, authentication flows, app store delivery.
Solar Monitoring App – Sunvigo GmbH
At Sunvigo I worked on a multi-platform mobile application that lets customers monitor their solar production and household consumption in real time, view their history and adjust key settings.
The goal was to make renewable energy usage feel transparent and approachable: clear graphs, simple controls and a robust backend that stayed responsive even with many customers online.
Tech: Flutter, Quarkus, REST APIs, authentication, monitoring and logging.
Speech Models for Low-Resource Languages
Together with colleagues from VU Amsterdam and partner organisations I built and evaluated an AI system that can learn a language word-by-word from native speakers, even when only a small number of speech samples are available.
As a proof of concept we supported Twi (Ghana) and Bambara (Mali), showing that communities don’t need massive datasets to start building useful speech technology in their own language.
Tech: Python, TensorFlow, PyTorch, speech processing, data collection tooling, evaluation pipelines.
Publication: ACM paper on low-resource speech technology
Medical Imaging Triage Models – VUmc Amsterdam
In collaboration with VU Amsterdam and VUmc I contributed to machine learning pipelines that analyse medical images such as MRI, CT and X-ray scans and flag cases that may deserve extra attention from clinicians.
The focus was not to replace radiologists, but to prioritise suspicious cases, reduce the chance of missed findings and support over-stretched hospital teams with consistent, explainable signals.
Tech: Python, deep learning frameworks, image preprocessing, training and evaluation workflows, clinical collaboration.
Psychological Coaching Website – Leonie Wittkamp
Designed and implemented a calm, welcoming brochure site for a psychological coach, focusing on clear structure, warm visuals and a frictionless contact flow for potential clients.
Tech: Jekyll on GitHub Pages, vanilla HTML, custom responsive CSS.
Author Website – “Solange wir uns hatten”
Built an author site for Leonie’s historical novel, with book presentation, events, contact and a subtle dark-gold visual theme inspired by the Orient Express motif in the story.
Tech: HTML, Bootstrap 5, custom CSS, GitHub Pages.