Lessons from the trenches of data & AI.
No hype, no fluff. Practical writing on what works — and what breaks — in production.
12 May 2026
SubstackA Minor Update, A Major Headache: GPT-5.4-Mini Broke Our App
A cautionary tale about blind model updates
Read on Substack18 May 2022
MediumBERT is eating your cash: quantization and ONNXRuntime to save 💰
Cutting Dutch GPT-2 inference costs by 2.4× with quantization and ONNXRuntime — without losing accuracy.
Read on Medium18 December 2020
MediumDutch GPT-2: Autoregressive Language Modelling on a budget
Finetuning English GPT-2 to Dutch on the OSCAR dataset, on a shoestring compute budget.
Read on Medium10 November 2020
MediumA Natural Language Processing look at the Belgian governmental agreement
Translation, summarization, and embeddings applied to Belgium's coalition agreement — where parties really land vs. their public stance.
Read on Medium22 May 2019
MediumGame of Thrones Twitter sentiment with Keras, Apache Beam, BigQuery and PubSub
An end-to-end streaming ML pipeline on GCP — Keras CNN classifier, Apache Beam, BigQuery and PubSub — over Game of Thrones tweets.
Read on Medium20 March 2019
MediumAdaptive and Cyclical Learning Rates using PyTorch
Comparing adaptive optimizers (Adam, RMSProp, Adadelta) with cyclical learning-rate scheduling on image classification.
Read on Medium