Machine Learning AppDeployment

3 Credits · Level: Graduate, Undergraduate

Machine learning application deployment bridges the gap between training a machine learning model and running it reliably in production. Many data scientists can build a model that functions well in a notebook. But far more interesting in industry is the question that immediately follows: how do I make this available to users, at scale, without it silently breaking over time? This class answers that question concretely, using a single semester-long project as the vehicle. By the end of the course, students will have built, deployed, and monitored a full ML-powered pipeline. The course is organized around a deliberate progression. The first part of the class establishes the engineering foundations that every subsequent week builds on: modern Python packaging, automated quality gates, probabilistic data structures, system architecture, and containerization. In the second part, we build the running service: a REST API, experiment tracking, distributed caching, and cloud deployment. The final part of the course focuses on making the service production-grade: reliability patterns, security, orchestration, and observability. Each week adds a concrete component to the project. The culminating activity of the course is the final project, where students will build and deploy their own sophisticated application that draws both from the skills we learn from the running example we study throughout the semester. The strategies discussed here are often components of state-of-the-art production systems. This class will help students build and deploy robust production-grade systems. Incrementally, students will construct production-grade, cloud-deployed AI systems, drawing from cutting-edge tools used in today’s AI ecosystems. Students are expected to be able to easily program in Python at an intermediate level and have strong prior exposure to deep learning or machine learning.