Home β€Ί Gen AI Fundamentals

🧠 Gen AI Fundamentals

Plain-English concepts & big-picture maps β€” the on-ramp to the CCA material

New to the Gen AI landscape? Start here. These pages build the mental models β€” who the players are, how the pieces fit, and what the jargon means β€” so the deep, exam-focused Claude Certified Architect (CCA) material lands more easily. Every page has diagrams and real-world examples.

Start here

βœ“ ready 🌌 The AI Universe β€” Who Does What The whole landscape on one page: the 6-layer stack, and how Anthropic, Claude, OpenAI, Google, Databricks, Azure AI Foundry, RAG, vector stores & MCP all fit together. βœ“ ready 🧩 Anatomy of a Gen AI Application The 11 building blocks of a real product β€” orchestration, RAG, memory, tools/MCP, caching, guardrails, eval β€” mapped to the CCA domains. βœ“ ready πŸ—ΊοΈ AI Engineering Blueprint β€” Concept β†’ Stack The architect's checklist: ~37 conceptual requirements (orchestration, eval, guardrails, memory, cost, injection, observability…) each mapped to open source, Databricks, Azure, AWS and GCP β€” plus Prompt vs Context vs Harness vs Loop vs Graph engineering, Tool vs MCP vs Skill vs Agent, and how to choose a memory store. βœ“ ready πŸ•ΈοΈ Ontology Builder β€” What, Who, Where Ontology vs taxonomy vs semantic layer, what an ontology contains, which formalism (OWL, SKOS, LinkML, property graph), who builds it, which store, PaaS vs on-prem β€” plus the life-sciences standards to adopt instead of building, and how these programmes usually fail. βœ“ ready πŸ›‘οΈ Implementing Guardrails β€” Catalogue to Code The 7-layer enforcement model and working Python for each control: injection, PII redaction, authorisation, output & citation validation, grounding, retries, loop budgets and action classes β€” with what to buy and what you must build. βœ“ ready 🧭 Stack Decisions β€” What Constrains What Seven layers from language to model, and the only two edges that are real constraints. Why "Claude" means two different things, which agent frameworks work through Kong to Azure OpenAI, and why ~216,000 permutations collapse to four decisions β€” only one of which is genuinely yours. βœ“ ready 🧰 AI Tooling Landscape Observability tools (and what OpenTelemetry's real alternatives are), AI gateways, orchestration & agent frameworks beyond LangGraph, and graph/memory frameworks β€” Graphiti, GraphRAG, Mem0 β€” with a default recommendation for each. βœ“ ready πŸ”¬ Observability & Eval β€” What Each Tool Offers Capability-by-capability matrices: Langfuse, Phoenix, MLflow, Opik, LangSmith, Braintrust, Datadog β€” and what "eval" actually means across unit, regression, RAG, agent, red-team and online. Plus the cloud-provider Rosetta stone: the same capability named in Databricks, Azure, AWS and GCP. βœ“ ready πŸšͺ AI Gateway β€” The Control Point The 14 capabilities, the open-source matrix (LiteLLM, Portkey, Envoy, Kong, Bifrost), and a real side-by-side of Mosaic AI Gateway vs Azure APIM vs the AWS pattern vs Apigee vs Cloudflare β€” with routing strategies, semantic caching's failure modes, MCP gateways, and what happens when it's down. βœ“ ready πŸ’‘ Bite-Sized Basics β€” Gen AI Q&A ~50 short questions with plain answers: how tokens are counted & why they differ by model, why pick one model/provider over another, RAG, agents, cost β€” plus quick cheat sheets.

Deeper explainers

βœ“ ready πŸ€– What's Inside an LLM β€” Transformers & Attention Tokens β†’ embeddings β†’ attention β†’ next token. How the Transformer works and why "next-token prediction" is so powerful. βœ“ ready βš™οΈ Training vs. Inference Building a model vs. using one β€” pre-training, SFT/RLHF, why the model doesn't learn from your prompts, and the adapt-a-model ladder.

Coming next

More single-concept explainers to add over time:

soonπŸ’Ύ Memory in Gen AIShort-term vs. long-term memory, the context window, and how apps remember across sessions. soon⚑ Caching, Latency & CostWhere the time and money go, and the levers β€” prompt caching, batching, model tiers β€” that move them. soonπŸ”Ž RAG & Vector Stores, in depthEmbeddings, similarity search, chunking, and building a retrieval pipeline end to end. soon🧭 Prompting, Agents & ToolsFrom a single prompt to an autonomous agent that plans, calls tools, and checks its own work.
Want one of the "coming next" pages prioritised? Just ask and it'll be built the same way.

Ready for the exam-focused deep dive? β†’ CCA-F Study Guide