The missing syllabus for shipping ML in the real world

What it does

applied-ml is a curated reading list: hundreds of papers and engineering blog posts from companies that have already shipped machine learning to production. The repo organizes them into 31 categories—data quality, feature stores, recommendation, search, ethics, team structure, even “fails”—so you can find how a specific problem was framed, what worked, what didn’t, and what ROI looked like.

The interesting bit

The curation is opinionated in a useful way. Entries are tagged by company and year, so you can trace how Uber’s feature store thinking evolved from 2017 to 2021, or watch Airbnb iterate on data engineering frameworks across four years. It’s less a bibliography and more a timeline of industrial consensus forming in real time.

Key highlights

• Covers the full lifecycle: data ingestion, discovery, feature stores, model management, A/B testing, infrastructure, and organizational design
• Heavy hitters represented: Google, Meta, Netflix, Uber, Airbnb, LinkedIn, Spotify, DoorDash, plus smaller shops like Monzo and Stitch Fix
• Explicitly includes failed approaches and lessons learned, not just success stories
• Companion repos for ml-surveys (research summaries) and applyingML (interviews and guides)
• Open to contributions with a CONTRIBUTING.md workflow

Caveats

• No original content: it’s purely links and categorization, so value depends on external sources staying live
• Some categories are sparser than others; depth varies by what companies have blogged about publicly

Verdict

Essential if you’re designing a production ML system and want to know how companies with similar scale solved (or stumbled on) the same problems. Skip it if you’re looking for code to clone or step-by-step implementation guides.