What happens once training stops: frozen weights, prefill vs decode, sampling, the KV cache, and the serving tricks that make inference faster without changing what the model knows.
How SFT and preference tuning (RLHF/DPO) turn a pre-trained next-token model into a helpful assistant: same training loop, different data, thin layer that shapes behavior not knowledge.
What happens inside the weight tensors as an LLM trains: random initialisation, the four-step training loop (forward, loss, backprop, optimizer), and what each component learns as a next-token model.
Netflix does not stream one file. Chunks, quality ladders, manifests, ABR decisions, buffering, seeking, and CDN caching, one segment at a time.
Resource APIs over HTTP/JSON versus function calls over HTTP/2 and protobuf. Stack placement, streaming, and when enterprises use both.
One HTTP application model, three wire contracts: text + TCP, multiplexed binary on TLS/TCP, and HTTP over QUIC on UDP. Head-of-line blocking, streams, and when each wins.
Layer 4 vs Layer 7 load balancing: what each sees on the wire, network vs application load balancers, algorithms, health checks, and when to use which.
Relational tables, document stores, key-value, graph, vector, and time-series: what each model is, when to use it, when not to, and the tradeoffs under the hood.
One HTTPS call from TCP open through TLS handshake, session keys, AEAD records, and teardown. Where asymmetric trust ends and symmetric bulk encryption begins.
HTTP is the application protocol for requests and responses. HTTPS is HTTP over TLS. Methods, status, headers, the TLS wrap, certificates, and which failures people mislabel as network or app.