// the find
mrdbourke/cs329s-ml-deployment-tutorial
Code and files to go along with CS329s machine learning model deployment tutorial.
A companion repo for Stanford's CS329S course that walks a Streamlit food-image classifier from a local script to a GCP-hosted deployment (AI Platform + App Engine, wrapped in Docker). It's for someone who has never shipped a model past a notebook and wants to see the whole path end to end, not a tool you install and reuse.
Explains the reasoning behind each piece (Dockerfile, Makefile, app.yaml) instead of just dropping commands, which is unusual discipline for a tutorial repo. It covers the full loop — train in model_training.ipynb, push to GCS, host on AI Platform, hit it from a Streamlit frontend, containerize, deploy to App Engine — so you see how the pieces connect rather than one disconnected demo. The README is honest about its own rough edges, calling out exactly where the app breaks (e.g. switching to Model 2 or 3), and it's backed by a full video walkthrough and slides.
It deploys to the legacy AI Platform Prediction product, which Google has been steering users off of toward Vertex AI for a few years now, so the console screenshots and gcloud commands in the README won't match current reality. Auth is done by downloading a long-lived service account JSON key, the exact credential pattern GCP's own current docs recommend against. There's no CI/CD or tests — GitHub Actions is floated in the 'Extensions' section as a someday-idea, not something actually wired up — so this is a one-time manual exercise rather than a pattern you'd carry into a real project. Last touched November 2022, so TensorFlow, Streamlit, and the GCP console have all shifted since; expect some translation rather than copy-paste.