Case Studies
2022 – 2023 Competitive Pricing Insights with AI
Via Valcon, I worked as data scientist for a large global retailer on the development of an AI-driven solution that generates competitive pricing insights through web scraping, state-of-the-art AI product matching, and a feedback loop for data-driven decision-making, which has now expanded globally.
2022 – 2023 Organizing a Full Week’s Data Science Workshop
At Valcon, I organized and enhanced a week-long data science workshop specifically tailored for our data consultants. The program was designed to span the entire data science lifecycle: starting with the business question, advancing through the development of a non-trivial data science solution, and concluding with the use of MLOPs tools and principles for continuous integration and deployment.
The workshop featured a series of talks delivered by internal experts, alongside with hands-on lab assignments to stimulate an interactive learning environment.
2020 – 2022 Algorithmic Food Authentication – Infrared Fingerprinting and Anomaly Detection in milk
I developed a quality assurance tool at Qlip, consisting of algorithms to detect and cluster anomalous milk samples using infrared fingerprinting. The methodology and findings on a large-scale dataset are published in an internationally recognized scientific journal. Beginning 2022, the QA tool was used in the context of a pilot project with several dairy processors in the Netherlands.
• Conceptualizing, developing, and testing of the algorithms
• Communication with business stakeholders using oral presentations and written reports
• First author of a scientific publication about the development process, findings, and implications
• Project leader for the pilot project
Key skills:
• Anomaly detection on high-dimensional chemometric data with seasonal profiles.
• Key techniques: data modeling and -reconstruction, multivariate distance metrics, generative cluster algorithms.
• Developing software to generate reports with text and visualizations to provide actionable insights about adulterated milk samples
• Leading a small team
• Communication with external and internal stakeholders
2021 – 2022 Development and Implementation of a Risk-Based Strategy for On-Site Grazing Inspections
Meadow milk holds greater economic value than its non-meadow counterpart. However, authentication of on-site grazing-related farming practices is expensive, time-consuming, and only provides snapshot information. We developed a simple yet effective selection strategy that continuously monitors all dairy farms and automatically selects farms for on-site inspections in a risk-based manner and in real-time. In 2022, the Dutch Grazing Foundation gave permission to use the selection strategy sector-wide. I presented our selection strategy at an international conference.
• Developing the selection strategy
• Conducting a large pre-registered empirical validation study of the selection strategy
• Communication with various stakeholders and business owners in written and verbal format
• Implementing the selection strategy
Key skills:
• Designing and planning a large empirical validation study
• Statistical analyses
• Running simulations
2020 – 2022 Data-Driven Quality Assurance: Predicting Milk and Blood Composition
At Qlip, I developed computational models using infrared spectroscopy for rapid and economical analysis of various chemical compounds in milk and blood. Additionally, I developed algorithms to detect measurement instabilities in infrared analyzers in real-time, and gave a talk about this novel methodology at an international conference.
• Developing and testing chemometric prediction models
• Planning data collection schemes
• Project lead: coordinating various stakeholders and processes from universities, veterinarians, laboratories, and logistics
Key skills:
• Laboratory statistics
• Optimized data collection to develop models with high dimensional data
• Project management
2020 – 2021 Optimizing Dairy Farm Inspections: Predictive Analytics for Risk-Based Selection
To ensure the integrity of dairy production conditions, including hygiene and animal welfare, it is necessary to conduct on-site farm inspections. Due to the sheer number of dairy producers, only a select few can be inspected annually, warranting a system that automatically chooses farms based on risk assessment. I have developed and implemented algorithms that leverage farm-related data, such as milk composition and past inspections, to forecast inspection outcomes, thereby streamlining the prioritization of farms for risk-based inspections.
• Developing, testing, and implementing the algorithms
• Communication with internal and external stakeholders
Key skills:
• Extracting useful information from highly diverse and noisy datasets
• Ensuring the model is fair and explaining to non-technical audiences the inner workings of the algorithms
• Key techniques: Gradient boosting, artificial neural networks