A Curated List of ML System Design Case Studies
An indexed table of 300+ engineering-blog case studies from 80+ companies on ML systems running in production, spanning 2017–2023. Maintained on GitHub by Engineer1999; ~10.9k stars / 1.7k forks. Each row carries company, industry, a few-word use-case tag, the article link and its year, so the collection is searchable by sector or by problem. That searchability is the actual product, since the writing belongs to the companies rather than to the list.
What the corpus looks like
Coverage skews to consumer-scale platforms, and clusters into a handful of recurring problems:
- Recommendation and personalization — Netflix, Pinterest, Spotify
- Fraud and risk — Stripe, PayPal, Nubank
- Logistics and ETA prediction — Uber, DoorDash, Lyft, Swiggy
- Search and ranking — LinkedIn, Etsy, Zillow
- Retail and marketplace — Walmart, Instacart, Etsy, Zillow
The industries that dominate are the ones where a small ranking improvement pays immediately, which is a fact about which ML systems get written about as much as which get built.
Why a link list earns a page
It is the only source in this spoke’s founding corpus describing systems that actually shipped, and it is evidence about the shape of the field: ml-system-design as practised is mostly ranking, matching and forecasting over behavioural data, not the model-architecture work the literature foregrounds. Set against ai-projects-fail-infrastructure-people, it is the surviving half of the same story: the systems that crossed the demo-to-production-gap and were written up afterwards, with the failures absent by construction.
Tier
T3 as an artifact — a pointer catalog with no original analysis, no stated inclusion criteria, and no maintenance guarantee; the newest entries are from 2023, so it is already stale as a picture of current practice. The sources it indexes are mostly T2 first-party engineering write-ups, which is where its value sits: it is a bibliography, and should be spent by following individual entries into this wiki, not cited as a claim in itself.
Survivorship warning: a company blog post about an ML system is a success story by definition. This corpus cannot tell you how often the same designs failed.
Related
ml-system-design · demo-to-production-gap · machine-learning · ai-projects-fail-infrastructure-people · synthesis