Applied AI Engineer · Taylor Corporation · Minnesota
I build the serving infrastructure, computer vision pipelines, and governance that move models out of notebooks and into daily operations. Five peer-reviewed papers on the research side, one Best Paper award.
Open-source models released
Hours of manual work removed by one pipeline
Peer-reviewed papers, one Best Paper award
Conference and guest talks delivered
01 / Selected work
Production systems at Taylor Corporation, plus the research that fed them.
Designed and deployed a vLLM serving stack on owned infrastructure, removing the company's dependency on third-party AI APIs. It now hosts fine-tuning on proprietary company data, work that could not legally or commercially leave the building.
A production pipeline that reads unstructured carrier documentation at full invoice volume, combining YOLOv11 detection, ResNet-50 classification, and a self-hosted Qwen LLM. Three model families in one path, running against real operational throughput.
An extraction engine that cut more than 530 hours of manual work. The interesting part was not the model. It was making the output trustworthy enough that a team stopped double-checking it.
Established the practice from nothing: model documentation, risk review before deployment, and post-deployment effectiveness audits. Someone has to be able to answer what a model does and whether it still works. Now that is written down.
A spatial-attention CNN that detects snow on sidewalks in real time, built for pedestrian safety and shipped to a phone as SnowWatch. Started as a thesis, ended as an Android app in private testing, with the mobile quantization and consent-flow problems that implies.
How far can prompt-based and inverse diffusion models carry a training set when the real data barely exists? Published in JMWAIS, and the award-winning CADSCOM paper came out of the same line of work.
02 / Stack
03 / Research
Computer vision, synthetic data, and generative model evaluation, mostly aimed at safety problems.
Full record on Google Scholar.
04 / Writing
Longer-form pieces on Medium, mostly about building computer vision systems from first principles.
05 / Speaking
Slides and details for each talk are on the highlights page.
06 / About
I am Dutch, and I now build AI systems in Minnesota. I earned my Master's in Data Science at Minnesota State University, Mankato, graduating top of my class, after a Bachelor's in ICT at HZ University of Applied Sciences with an Erasmus+ minor in Computer Science in Austria.
My thesis put a spatial-attention network on a phone to detect snow on sidewalks in real time. That set the pattern for everything since: take a model that works in a paper and make it survive contact with production. At Taylor Corporation that means owning the serving infrastructure, the vision pipelines running against real invoice volume, and the governance that keeps both accountable.
Outside of that I compete in data hackathons like Data Derby and MUDAC, and speak at conferences, usually about making a hard architecture legible to people meeting it for the first time.
07 / Contact
Open to conversations about production LLM systems, computer vision, and speaking engagements. I read everything that comes through here.