Skip to content

GenAI Topics

Last reviewed: October 2026

The building blocks of generative and agentic AI, from how models work through to running them reliably in production. Each topic page has diagrams, the trade-offs that matter, and an interview deep dive.

Who it's for: engineers preparing for senior, staff or principal AI, ML platform or forward-deployed roles who need to explain why a design works, not just which library to import. If you'd rather build something first, start with the Setup Guides and come back here for the theory.

Learn → Build → Ship → Grow

Every topic an AI engineer is expected to know, in the order to learn it, and the page here that covers it. Learn one topic a week and build something small with each one.

flowchart LR
    L[Learn<br/>how LLMs work, prompts,<br/>context, models] --> B[Build<br/>embeddings, vector DBs, RAG,<br/>agents and frameworks]
    B --> S[Ship<br/>MCP, safety, evals,<br/>observability, cost, ops]
    S --> G[Grow<br/>projects, interviews,<br/>economics, trends]

Learn

Topic Read
How LLMs work: tokens, attention, sampling, temperature LLM Fundamentals
Prompt engineering: few-shot, chain of thought, structured output Prompt Engineering
Context engineering: memory, retrieval, what goes in the window Context Engineering
Models and model choice: open vs closed, cost, latency, quality LLM Fundamentals · AI economics

Build

Topic Read
Embeddings and embedding models Embedding Models
Vector databases: indexing, chunking, similarity search Vector DB
RAG: retrieval, re-ranking, hybrid search, RAG vs fine-tuning RAG · Retrieval Tuning
Knowledge graphs and GraphRAG Graph DB
Agents: tool calling, planning, reflection, memory, multi-agent Agent architecture · Agent Engineering · Building Agents · 7 diagrams
Frameworks and platforms LangChain · LangGraph · Bedrock · AgentCore

Ship

Topic Read
MCP and agent-to-agent protocols MCP · A2A
Safety: prompt injection, guardrails, output validation, access AI Security · Guardrails
Evals and observability: tracing, regression tests, LLM judges Observability & Eval
How AI work differs from normal software: SDLC vs the AI lifecycle SDLC vs AI lifecycle
Production: rate limits, caching, queues, scaling, incidents LLMOps · Reliability · Kubernetes · DevOps for AI
Cost control Cost Optimization

Grow

Topic Read
Portfolio projects 7 AI projects
Courses, if you need the basics first Free courses, in order
The business side AI economics · Trends & market
Interview preparation OfferReady app: plan, practice and mock interviews for your job

Not covered here yet: multimodal AI (vision, speech, image and video generation) and AI coding tools. Their official docs are the best place to start.

Short on time? Start with the pictures

Agentic AI in 7 diagrams: harness, context, tools, memory, guardrails, evals and tracing, each as one flow diagram and five steps. New to AI? Free courses, in order lists the few you need, stage by stage. Want the business view? AI economics in 2026: who gets paid, why inference is the bill that grows, and the interview questions it brings. Ready to build? 7 AI projects that get you interviews.

1. Foundations

  • LLM Fundamentals


    Tokens, attention, sampling, context windows, and when to use fine-tuning versus RAG.

    Open

  • Prompt Engineering


    Instructions, few-shot examples, reasoning strategies and structured output.

    Open

  • Context Engineering


    Deciding what goes into the context window: assembly, memory, compaction and token budgets.

    Open

  • Embedding Models


    How text becomes vectors, and how to choose and use embedding models for search and RAG.

    Open

2. Retrieval and knowledge

  • RAG


    Retrieval-augmented generation end to end: indexing, retrieval, grounding and citations.

    Open

  • Vector Databases


    How similarity search works, ANN indexes, distance metrics and choosing a store.

    Open

  • Retrieval Tuning


    Top-K, metadata filtering and reranking: the levers with the biggest effect on RAG quality.

    Open

  • Graph DB & GraphRAG


    Nodes, relationships, knowledge graphs, and retrieval that follows connections.

    Open

3. Orchestration and agents

  • LangChain


    Composing prompts, models, retrievers and tools into LLM applications.

    Open

  • LangGraph


    Agent workflows as stateful graphs, with cycles, branching and human-in-the-loop.

    Open

  • MCP


    Model Context Protocol, the open standard that connects LLM hosts to tools and data servers.

    Open

  • A2A


    Agent-to-agent communication for delegating and coordinating across teams of agents.

    Open

  • Agent Engineering


    A survey of building agents that reason, use tools and act reliably.

    Open

  • Agent Principles & Patterns


    A code-level deep dive into agent principles, control-flow patterns and the anti-patterns to avoid.

    Open

  • AgentCore


    Amazon Bedrock AgentCore: runtime, memory, gateway, identity and observability for production agents.

    Open

  • Amazon Bedrock


    AWS's managed multi-provider model API, with knowledge bases, agents and guardrails.

    Open

4. Production

  • Observability & Evaluation


    Tracing, evaluation sets, LLM-as-judge and production monitoring.

    Open

  • LLMOps / Deployment


    Serving, scaling, caching, versioning and the lifecycle of models and prompts.

    Open

  • Cost Optimization


    A practical playbook for cutting token and call costs without losing quality.

    Open

  • Reliability & Distributed Systems


    Timeouts, retries, circuit breakers and the failure modes specific to AI systems.

    Open

  • DevOps for AI


    Version control, infrastructure as code, CI/CD pipelines and deployment strategies for AI systems.

    Open

  • Kubernetes & Containers


    Running model serving and GPU workloads: images, cold starts, autoscaling.

    Open

5. Landscape

  • GenAI Trends & Market


    A dated snapshot of where the field is, so your answers reflect today rather than 2023.

    Open

Practise what you read