ML Engineer · Amsterdam

Wouter Wijffels

I build ML systems that run in production — churn models deployed across 5 countries, LLM pipelines processing a million+ interactions, monitoring infrastructure that catches problems before they become incidents. Most of my time goes into the boring-but-hard parts: data contracts, multi-market deployment, keeping models honest over time.

15M
customers
5
countries
15
models in prod
400+
pipelines

LLM Summarization & Classification

Near real-time pipeline on AWS Bedrock: conversations ingested → PII masked → summarized and classified → auto-filed in CRM. 1.3M interactions per year.

1.3M interactions/year
10% wrap-up time savings
AWS Bedrock, Lambda, SageMaker, MLflow

International Churn Model Deployment

Single XGBoost model deployed to 5 countries via S3 distribution. Each country has its own data warehouse and marketing automation — all read from the same prediction store.

5 countries
A/B tests running in all markets
SageMaker, Snowflake, S3, XGBoost

MLOps Monitoring Platform

Centralized observability for 15 models and 400+ pipelines. Drift detection uses statistical process control — not threshold alerts — which cuts false positives significantly.

15 models, 400+ pipelines
Early drift detection via SPC
Terraform, Lambda, MLflow, Prometheus

Content Personalization Engine

LLM reads marketing automation group selections and generates personalized campaign prompts per segment. Pilot running.

In pilot
Snowflake, Cortex, AWS Bedrock, Lambda

A/B Test Uplift Prediction

Predicts expected uplift per segment before a campaign launches, using historical experiment data. Replaced intuition-based targeting with model-driven decisions.

~35% uplift improvement
Python, Polars, Snowflake, Scikit-learn

Semantic Search for Procurement

Hierarchical item clustering with supplier price comparison. Drag in a price offer PDF and instantly find comparable products from other suppliers.

In evaluation
OpenAI Embeddings, FAISS, Streamlit

Procurement Analytics

Margin tracking per supplier, product, and category against webscraped competitor prices. Built when Picnic left its buying cooperative and started negotiating directly.

~5% margin improvement
Python, Snowflake, Google Sheets

Supply Forecasting System

Hybrid time series + regression forecasting for brewery supply. COVID-19 signals from Our World in Data used as regression features. Confidence intervals fed into stochastic allocation.

8.5/10 · scaled to glass & carton
Prophet, Exponential Smoothing, Python, SQL

Commodity & Gas Index Mapping

NLP mapping of public gas and commodity indices to supplier cost structures — built to verify whether supplier price increase claims during the Ukraine war were justified.

POC
Python, NLP, Snowflake, Commodities APIs

Email Scraper — 3PL Stock Updates

One-button local app that parses HTML stock update emails from the logistics partner and outputs structured Excel files per warehouse. Rolled out to 4 country operations teams.

BeautifulSoup, Gmail API, Python, openpyxl

Stack

AI & ML
PyTorchHuggingFaceAWS BedrockPolarsScikit-learn
Data
SnowflakeBigQuerySQLAirflow
Infrastructure
SageMakerMLflowDockerTerraformGitLab CI/CD

Background

MSc Transport, Infrastructure & Logistics — Cum Laude

Delft University of Technology

BSc Industrial Engineering & Management

University of Groningen

Started with operations research and logistics optimisation, moved into data science when the tooling caught up with the problems. Most of my work since has been on the gap between a working notebook and a model that reliably serves predictions in production — deployment patterns, monitoring, multi-country infrastructure.