Why AI Infrastructure Is a Career Move Right Now ↗
Learn why reliable model serving sits at the intersection of software, distributed systems, performance engineering, and business outcomes.
LLM Infrastructure From First Principles
A practical path from models, tokens, and GPU sizing to production inference, intelligent routing, and Kubernetes.
How ChatGPT and Large Language Models Work ↗
Follow a request from text through tokens, transformer computation, probability sampling, and the production systems around an LLM.
How AI Models Are Stored and Why Hardware Limits Matter ↗
Understand checkpoints, precision, runtime memory, model provenance, and why many large models exceed a typical laptop without oversimplifying the tradeoffs.
GPU Sizing for LLM Inference Without Guesswork ↗
Size inference capacity using weight memory, active-token demand, and service objectives while accounting for GPU compute, VRAM, and bandwidth.
From PyTorch Weights to a vLLM Model Server ↗
Understand the boundary between model code and a production inference server, then design a controlled path from artifact to API.
Prefill, Decode, TTFT, and TPOT Explained ↗
Connect the two phases of language-model inference to user-visible latency, capacity tests, and practical performance diagnosis.
KV Cache, Prefix Caching, and Continuous Batching ↗
Learn how LLM servers reuse attention state, combine active requests, protect tenant boundaries, and balance latency against throughput.
Distributed LLM Inference, Sharding, and Routing ↗
Choose the right parallelism and route requests using real workload cost, cache locality, and failure boundaries instead of replica counts alone.
Running llm-d on Kubernetes in Production ↗
Turn the series into a deployment plan using llm-d routing, Well-Lit Paths, Kubernetes workload choices, observability, and scoped hands-on labs.