MLOps for the rest of us- A poor man's guide to putting models in production
summary
Argues small teams can reliably ship models without a full ML platform, showing concrete tools and numbers: only 22% of companies have models in production, use lightweight stacks like FastAPI and Pydantic, package with Docker + GitHub Actions and manage deps with pip-tools. Shows data tricks (embeddings, nearest‑neighbor, fine‑tuning) with an example of 223 samples → F1 0.7. Demonstrates a working containerized API, saves inputs/outputs to BigQuery, and exposes a live drift dashboard at /routing so monitoring and deployment actually work.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:30Why not like LinkedIn
- 3:36MLOps value over tools
- 7:08People and skepticism in ML
- 10:17Data scarcity techniques
- 12:26Lightweight project setups
- 13:49Testing and CI/CD
- 17:19Containerized models to production
- 18:45Cloud service selection
- 21:14Three monitoring types
- 22:55Simple ML platform with cookie cutters
- 24:05Key takeaways
related talks
50mBuilding an MLOps strategy from the ground up
25mBuilding Scalable MLOps Pipelines That Actually Work
47mMachine Learning in Production: An Intro to MLOps.
13mMLOps at Scale
45mThe Culture, Architecture and Technology of MLOps
34mMLOPS 101: Tips, Tricks and Best Practices
01h 13mPractical MLOps- Do we need all the things
40mA Hitchhiker Guide to MLOps