Learning AI engineering in public

Learn AI engineering,
one topic at a time.

I write up one AI engineering topic at a time: what I understood and the resources I used. Follow along as it grows.

Small, focused and explained from first principles, so you can read it over coffee.

The papers, videos and posts I actually learned from, linked on every page.

There is no fixed syllabus. The next topic is whatever is worth understanding next.

Attention
How a token decides which other tokens matter, from Q, K, V to multi-head attention.
KV Cache
How KV Cache speeds up LLM inference by reusing Keys and Values, the GPU memory it costs, and how PagedAttention, GQA and MQA tame it.
Speculative Decoding
How a small draft model proposes tokens and a large target model verifies them together, cutting the sequential steps of LLM generation.
Quantization
How storing model weights in fewer bits cuts memory, bandwidth and cost, and what it trades away in numerical precision.
Fine Tuning
Updating a pretrained model's weights on a task-specific dataset to change its behavior, and when that beats prompting or RAG.
Mixture of Experts
How MoE models use a router to activate only a subset of experts per token, trading engineering complexity for more capacity at a lower compute cost.

Agent Loop
The loop that sends context to a model, executes the tool it requests, and feeds the result back until the model returns a final answer.
Tool Calling
How a model requests a function, your application executes it, and the result goes back to the model, plus the failure cases to handle.
Context Engineering
How to select, organize and update the information an LLM receives, from recent-message selection to summarization, retrieval and structured state.
Agent Memory
How agents store, retrieve, update and forget information across conversations, and how short-term, long-term, episodic and semantic memory fit with context engineering.
Planning and Task Decomposition
How agents break a goal into tasks with dependencies and status, execute them in order, verify results, and replan when a task fails.
Model Context Protocol
What MCP is, how hosts, clients and servers work together, how tools are discovered and called, and when to use it instead of plain function calling.
Human-in-the-Loop (HITL)
How an agent pauses at a checkpoint, waits for a human to approve, reject or change an action, and resumes, plus a runnable approval gate for a refund agent.
Agent Workflows and Multi-Agent Systems
How to structure a task as sequential, parallel, conditional or iterative steps, when to split work across multiple agents, and how a coordinator passes information between them, plus a runnable research pipeline.

I share what I'm learning as I go. If you're on the same path, or spot something I got wrong, I'm@ratishtwts on X.