300+ ML war stories from the trenches of Netflix, Stripe, and Uber

What it does This repository is essentially a well-organized reading list: 300+ case studies from 80+ companies, catalogued by industry and ML use case. Each entry links to original engineering blogs or papers where teams explain what they built, why, and how it fell over in production.

The interesting bit The value is in the curation, not the content itself. Someone did the tedious work of reading hundreds of corporate engineering blogs, extracting the ones that discuss actual system design (not just model architecture), and sorting them by whether you care about fraud detection, recommendation, or LLM incident management. The README includes a giant table with company, industry, a five-word description, and the original source link.

Key highlights

• Covers Stripe’s fraud radar, Uber’s airport demand forecasting, GitHub Copilot’s LLM plumbing, and Airbnb’s guest-interest modeling
• Organized by practical categories: computer vision, NLP, recommender systems, search/ranking, fraud detection
• Sources are explicitly “in-house” engineering write-ups, not third-party analysis or marketing gloss
• Includes non-obvious applications: LinkedIn payment gateway selection, Honeycomb natural-language query generation, Zillow’s “Neural Zestimate”
• Roughly 10,000 GitHub stars suggest the audience is mostly interview-prepers and system-design tourists

Caveats

• The repository itself contains no original analysis, just links and brief descriptions; you still have to read the source material
• A large portion of the README is actually promotion for horizonx.live, an unrelated research platform with a laundry list of “upcoming features”
• The table formatting in the README appears truncated in places, and some descriptions are vague (“Recommend items to order” tells you little about the system)

Verdict Worth bookmarking if you’re preparing for ML system design interviews or need to quickly find precedent for a problem you’re facing. Skip it if you want depth without clicking through to original sources — this is a map, not the territory.