AI Mastery: From Beginner to Expert
A complete, offline, self-paced curriculum for developers who want to understand AI deeply and use it every day, from writing their first prompt to designing agents, MCP servers, evaluation pipelines and AI-native systems. Examples are in Scala and Python.
Why this exists
Most AI material falls into one of two traps. It is either shallow ("10 prompts that will change your life") or it jumps straight into maths and research papers. This curriculum climbs the whole mountain one step at a time. Each level builds on the previous one and ends with hands-on labs and a checkpoint, so you know you are ready to move up.
The goal is not only to "use AI tools". By the end you should be able to:
- Explain how modern AI models work, why they fail, and how to get reliable results from them.
- Make AI a real multiplier in your IDE: agents, instructions, skills, MCP servers and review workflows.
- Build AI features into products: APIs, RAG, agents, tools, evaluations, guardrails, cost control.
- Use AI to learn any new language, framework or domain much faster, without becoming dependent on it.
- Make sound decisions about security, privacy, IP and responsible use.
- Go deep on the internals (transformers, training, inference), read research, and lead AI adoption in a team.
The five levels
See the Curriculum Roadmap for every topic in every level, across 20 learning tracks.
Start here: Level 1 Foundations
01 · What Is AI?
AI, ML, deep learning, generative AI and LLMs. How they relate and a short history.
02 · How LLMs Work
Tokens, embeddings, attention, training, inference, temperature, context windows.
03 · Prompting
Anatomy of a great prompt, core patterns, iteration, and developer-specific templates.
04 · Context Engineering
The skill that matters most: giving the model the right information at the right time.
05 · Tools Landscape
Chat assistants, IDE assistants, CLI agents, local models, frameworks. A map of it all.
06 · AI in the IDE
VS Code, IntelliJ IDEA and Xcode compared. Completion, chat, inline edit and agent modes.
07 · Instructions & Skills
AGENTS.md, rules, prompt files, skills, custom agents, hooks. Teach the AI your way.
08 · Agents 101
The agent loop, tools, planning, memory, autonomy levels and how to supervise agents.
09 · MCP 101
The Model Context Protocol: the "USB-C for AI tools". Architecture, config, safety.
10 · Scala with AI
Metals, sbt, scala-cli, Scala 3 idioms, testing, Spark, and Scala-specific prompting.
11 · Python with AI
uv, ruff, pytest, typing, notebooks, and your first LLM API call.
12 · AI Across IT
DevOps, SRE, QA, data, security, docs, architecture and project management.
13 · Learn Faster
Use AI as a tutor, coach and sparring partner to grow faster in anything.
14 · Safety & Ethics
Hallucinations, privacy, secrets, IP, prompt injection, bias, and verification habits.
15 · Hands-on Labs
12 guided exercises in Scala and Python. You learn by doing.
16 · Checkpoint
An interactive quiz. Score 80%+ and you are ready for Level 2.
Continue: Level 2 Practitioner
L2 01 · Prompt Patterns
Prompt chains, generator–critic, self-consistency, routing, prompts as tested code.
L2 02 · Structured Output
JSON Schema, circe and pydantic decoding, validation and a repair loop.
L2 03 · Calling LLM APIs
A production-shaped client: config, retries, back-off, streaming, tokens and cost.
L2 04 · Tool Calling
Build the core of an agent: a Store Assistant that checks stock and orders.
L2 05 · Embeddings
Semantic product search, cosine similarity, hybrid search with RRF.
L2 06 · RAG
A policy assistant that answers only from your docs, with checked citations.
L2 07 · Agentic Coding
Specs, test-first loops, refactors, Scala 2→3 migration, parallel agents.
L2 08 · Skills & Agents
Descriptions that trigger, scripts, trigger tests, planner/reviewer roles.
L2 09 · MCP Servers
Choosing, configuring and securing servers; Metals MCP in depth.
L2 10 · Testing & Review
Property-based tests, mutation testing, targeted AI reviews.
L2 11 · Local Models
Ollama, LM Studio, llama.cpp, MLX; sizing and quantisation.
L2 12 · Data Engineering
Safe text-to-SQL, Spark plans, data-quality checks, LLMs in pipelines.
L2 13 · Labs
12 hands-on labs that build the capstone pieces.
L2 14 · Capstone
The complete Store Assistant: tools + RAG + search, with a golden set.
L2 15 · Checkpoint
20 questions. Score 80%+ to move to Level 3.
Go deeper: Level 3 Advanced
L3 01 · Build MCP Servers
FastMCP in Python, the JVM route from Scala, the protocol by hand, testing, packaging.
L3 02 · Agents from Scratch
The loop done properly: state, memory, error recovery, limits and budgets.
L3 03 · Multi-Agent
Orchestrator–worker, author–critic, artefacts over conversations, A2A.
L3 04 · Evals
Golden sets, code graders, LLM-as-judge pitfalls, trajectory evals, CI.
L3 05 · Advanced RAG
Hybrid BM25+vector, RRF, re-rankers, query rewriting, GraphRAG, agentic retrieval.
L3 06 · Fine-tuning
When (not) to fine-tune, SFT data, LoRA/QLoRA, distillation, honest eval.
L3 07 · LLMOps
Tracing, caching, routing, cost control, guardrails, canary deploys.
L3 08 · AI Security
Prompt injection, tool validation, MCP supply-chain risks, OWASP Top 10.
L3 09 · Labs
8 bigger labs: ship an MCP server, a durable agent, evals, red-teaming.
L3 10 · Capstone
A production agent: own MCP server, eval suite, observability, security review.
L3 11 · Checkpoint
20 questions. Score 80%+ to move to Level 4.
Open the box: Level 4 Expert
L4 01 · The Maths
Vectors, matmul, softmax, cross-entropy, gradient descent — all in runnable code.
L4 02 · Transformers
Attention from scratch, multi-head, residuals, positional encodings, a tiny GPT.
L4 03 · Training
Pre-training, SFT, RLHF vs DPO, RLVR — how a completer becomes an assistant.
L4 04 · Inference
KV cache, continuous batching, paged attention, speculative decoding, serving.
L4 05 · Reasoning
Test-time compute, verifiers, effort dials — when "thinking" pays off.
L4 06 · Architecture
AI gateways, build vs buy, platform teams, cost and reliability models.
L4 07 · Governance
EU AI Act tiers, model risk management, documentation, paved paths.
L4 08 · Labs
7 labs: implement attention, train a tiny GPT, benchmark serving.
L4 09 · Capstone
Train, quantise, serve and benchmark a tiny GPT + design doc.
L4 10 · Checkpoint
20 questions. Score 80%+ to move to Level 5.
The frontier: Level 5 Master
L5 01 · Read Research
Three-pass reading, falsifiable claims, reproductions and ablations.
L5 02 · Interpretability
Superposition, probes, SAEs, steering; alignment and reward hacking.
L5 03 · On-device AI
Apple silicon, Core ML, MLX, the ANE, hybrid on-device/cloud design.
L5 04 · Multimodal
VLMs, speech (ASR/TTS), video, diffusion — uses, costs, failure modes.
L5 05 · Lead Adoption
Pilots, honest productivity measurement, training, policy that works.
L5 06 · Labs
5 open-ended labs that produce real artefacts.
L5 07 · Capstone
Publish a speciality project: blog, tool, reproduction, talk or paper.
L5 08 · Checkpoint
15 questions — the last checkpoint. Then publish your capstone.
How to use these docs
- Go in order within a level. Pages build on each other. Each one starts with learning objectives and ends with key takeaways.
- Keep an AI assistant open next to the docs. Every page has "Try it" boxes. Do them, because reading alone will not make you an expert.
- Mark pages complete. Use the button at the bottom of each page. Your progress shows in the sidebar and is saved in this browser.
- Search everything. Press / or Ctrl+K. Unknown word? Check the Glossary.
- Do the labs and the checkpoint before moving up a level.
- Plan time. Level 1 takes about 15–25 hours including labs. Spread it over 2–4 weeks and practise daily.
Page features
| You will see | Meaning |
|---|---|
| Level 1 25 min | Difficulty level and estimated reading time (labs extra). |
| Note box | Important background or nuance. |
| Tip box | A practical habit used by experienced practitioners. |
| Warning box | A common mistake or risk. |
| Analogy box | An everyday comparison that makes the idea stick. |
| Try it box | A small exercise to do now with any AI assistant. |
| Deep dive (expandable) | Optional extra detail. Skip it on first reading if you like. |
Sharing
The whole site is plain HTML, CSS and JavaScript. It needs no internet, no server and no installation. To share it,
zip the folder and send it. The recipient unzips it and double-clicks index.html.