Our MLOps Story: Production-Grade Machine Learning for Twelve Brands
Things we learned building an MLOps platform with limited means at DPG Media: SageMaker, MLflow, Terraform, and a team of two serving a dozen brands.
// 10 posts
Notes on machine learning engineering, cloud, NLP, and the occasional side project. Oldest archives go back to 2016.
Things we learned building an MLOps platform with limited means at DPG Media: SageMaker, MLflow, Terraform, and a team of two serving a dozen brands.
The ML Engineer title went mainstream, but MLOps never quite consolidated. Notes on the tooling sprawl and the platform we built at DPG Media.
A fun exercise in text analytics on job postings and a resume, starring tf-idf and t-SNE, to find which data role actually suits my profile best.
I collected the job titles of the people I started my Computer Science degree with, to see where a cohort actually lands years later.
I'm a terrible birder, so I set out to scrape 60,000 labelled bird photos and build a CNN classifier, with a Raspberry Pi as the eventual home.
I passed the GCP Professional Data Engineer exam. A short note on the exam, why I took it, and how I studied for it.
A free-format project on the SurfSara Hathi cluster: running Spark over 1.73 billion crawled web pages to build word clouds around kayaking.
Poking at a Kaggle housing dataset in Spark: loading 292 columns, running SQL over it, and hunting for the Kremlin in the data.
A hello-world walk through Hadoop and HDFS: setting up a pseudo-distributed cluster and running the classic MapReduce word count on the complete Shakespeare.
The Dutch road authority put live gritter-truck positions behind an unadvertised API, so I logged it over time to map where they actually drive.