Notes

Build a small causal next-word model, inspect every LSTM state and tensor shape, train it in PyTorch, and understand why bidirectional LSTMs do not fit this training objective.

Build a small positive-or-negative sentence classifier, follow every tensor through tokenization, embeddings, recurrent memory, and the classifier, then train and evaluate it in PyTorch.

A small, visual introduction to recurrent neural networks: generate slow and fast sine waves, follow every tensor shape through a hand-built RNN, train it with backpropagation through time, and evaluate it.

How Docker and Kubernetes keep configuration out of the image, the two ways config and secrets reach a container, why files update live but environment variables do not, and how to actually secure Secrets.

How Docker keeps container data on one machine, how Kubernetes provides persistent and distributed storage across a cluster, and how to choose the right option for databases and machine learning.

How Docker builds containers from Linux features, how Kubernetes runs them across many machines, and how the two fit together.

How Raft keeps a group of computers in agreement: leader election, log replication, and its role inside etcd and Kubernetes.

FastAPI essentials: routing, validation, and building a clean modern API in Python.

A working reference for ChromaDB: collections, queries, filters, and persistence.

A working reference for Faiss: index types, trade-offs, and the code to use them.