How you really get your data science models into production the cool way!
summary
Builds an API-first microservices pipeline that actually serves models: a Keras MLP on the MNIST dataset reaches ~98% accuracy, model artifacts go into Redis as a cache, and a lightweight Flask API plus a Sinatra sketchpad web app consumes predictions. Uses pair programming, TDD for production code, Concourse CI pipelines and Cloud Foundry (PCF/PCF Dev) for deploys. Demo runs but shows real-world quirks (handwritten 4s often misclassified as 8), illustrating what was built, how it’s tested, and how it’s deployed.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:48Unused models reality
- 3:35Models rarely reach production
- 4:57API-first model deployment
- 7:36Demo setup MNIST
- 15:33Exploration to production
- 16:36Production-ready services
- 18:46Deploying Cloud Foundry
- 21:59CI/CD pipeline
- 24:07Live demo sketchpad
- 26:51Key takeaways
- 29:46Test-driven development
- 30:07Q&A
related talks
35mGetting Machine Learning Models Ready For Production Using Python
32mDeploying Python models to production
35mHow to deploy machine learning models to production (frequently and safely)
17mR & Python playing nice, in Production
46mDev Ops meets Data Science Taking models from prototype to production with Docker
42mDeploying ML models and all the things that go wrong
46mProduction Model Deployment
26mMonitoring models in production