MLOps and AI systems in production From the experiment notebook to a system that runs for years: reproducible training pipelines and experiment tracking, monitoring models and drift, then designing AI systems at scale and securing them.
Lessons 1 Training pipelines and experiment tracking From a messy Jupyter notebook to a reproducible training pipeline: versioning data and code, fixing seeds, tracking experiments and their metrics, and a model registry. 2 Monitoring models and drift A model degrades silently after deployment: data drift and concept drift, how to measure it without labels, when to retrain, and safe rollouts with shadow and canary deploys. 3 Designing and securing AI systems An AI system at scale: the gateway and response caching, routing between models by cost, limits and queues, continuous evaluation, and securing language models against injection and data leaks.