Syllabus

Your curriculum

Six months full-time, up to nine part-time. A seven-week shared core, then two tracks 34 modules and 132 topics in total, ending in 17 things you can put in a portfolio.

Join The ProgramTry Free Course

Shared Core

Weeks 1–7 · everyone

ML Engineering

Train the models

AI Engineering

Build with models

Forward Deployed

Ship with users

Advanced systems · capstone

All three tracks reconnect

Everyone starts here

Shared Core · Weeks 1–7

Whichever track you take, you need the same foundation: code you do not have to think about, data you can move around, enough statistics to know when a result is real, and a development environment with intelligence built into it.

Module I

Programming Foundations

Weeks 1–3

PythonThe language properly — types, control flow, functions, modules, and the standard library you will actually reach for.
SQLSelects, joins, aggregation and window functions. Most real data still lives in a table someone else owns.
Git and GitHubBranches, pull requests, review and history. The skill that separates a script from a codebase.
APIs and HTTPRequests, responses, status codes, authentication and rate limits — the shape of every service you will build against.
PROJECT

A command-line tool that pulls from a public API, stores results, and ships with a README someone else can follow.

Module II

Working With Data

Week 4

PandasLoading, reshaping, joining and grouping. The daily work of getting data into a state worth analysing.
Cleaning and missingnessBad rows, wrong types, absent values, and deciding what to do about each without quietly changing the answer.
ExplorationLooking at a dataset before modelling it, and noticing the thing that would have ruined the model.

Module III

Statistics and ML Intuition

Weeks 5–6

Statistics essentialsDistributions, sampling, inference and validation — the discipline of knowing when a result is real.
ML fundamentalsWhat learning from data actually means: features, targets, fitting, generalisation, and why a model can be right and useless.
Neural network intuitionLayers, weights and gradient descent, built up from a single neuron. No frameworks yet.
EmbeddingsMeaning as coordinates. The idea that carries across all three tracks — nearest neighbours here become retrieval later.
PROJECT

An analysis of a real dataset that states a claim, tests it, and says plainly where the evidence runs out.

Module IV

AI-Assisted Development

Week 7

CursorAutocomplete through to agent mode, rules files, codebase context and background agents.
Claude CodeRepository understanding, planning, multi-file edits, terminal work, tests and Git.
Wispr FlowVoice-driven development: dictating requirements, debugging out loud, editing by speaking.
Working with intelligence in the loopNot a tour of tools. How an engineer actually works when the editor can reason — what to delegate, what to check, and what never to hand over.

Then it forks

Three tracks. Pick yours.

All three run seventeen weeks and reconnect for the final two weeks. They are not sealed off from one another — you may take electives across the other paths before the shared capstone.

For those who want to ship AI products using models, agents, retrieval, voice, tools and evaluation.

Module I

Working With LLMs

Weeks 8–9

Tokens and context windowsWhat the model actually reads, what it costs, and why long conversations get expensive.
Model familiesOpenAI, Claude and Gemini side by side — the real differences, and reasoning models as their own category.
System promptsThe instructions sent on every single turn, and what belongs in them.
Structured outputsGetting JSON you can rely on instead of prose you have to parse.
Streaming and multimodal inputToken-by-token responses, and passing images, audio and documents in.
8 more modules

Register to unlock the rest

An account opens every module across all three tracks, the project briefs and the reading list. It takes a minute and costs nothing.

Unlock The Full Curriculum

All three tracks, together

Advanced AI Systems · Weeks 25–26

The tracks are not separate universes. An AI engineer eventually needs deeper model intuition; an ML engineer eventually needs to ship their model inside a product; a forward deployed engineer needs enough of both to solve a customer workflow end to end. These weeks are shared, and you may take electives from either of the other paths.

  • Models inside agents
  • Fine-tuning for retrieval
  • Cross-track electives
  • Scope and brief
  • Build and evaluate
  • Review

What you walk out with

17 things for your portfolio

Every one is shippable, and every one names what it proves to somebody reading your CV. That last part is what most course pages leave out.

ML Engineering · Training ModelsFlagship

A fine-tuned model that beats the baseline

Adapted to your task, measured against the off-the-shelf version, with the numbers to prove the gap.

Proves You can improve a model deliberately rather than hopefully.

  • Fine-tuning
  • Transformers
AI Engineering · Retrieval-Augmented GenerationFlagship

AI over a company knowledge base

Answers from real documents with citations — and refuses when the answer genuinely is not in them.

Proves You can build retrieval that is honest about what it does not know.

  • RAG
  • Vector search
  • Reranking
AI Engineering · AgentsFlagship

Your own MCP server, and an agent that uses it

A server you wrote, plus an agent completing a multi-step task with human approval on the irreversible step.

Proves You understand agent architecture from the inside.

  • MCP
  • Tool calling
  • Agents
AI Engineering · Voice AIFlagship

An AI receptionist that answers a real phone call

Picks up an actual number, knows your documents, handles interruptions, and takes a message when it cannot help.

Proves You can build something a stranger can use without instructions.

  • ElevenLabs
  • Telephony
  • Latency
Forward Deployed Engineering · Applied AI WorkflowsFlagship

An AI-assisted operational workflow

Uses domain context and tools to complete useful work, with human approval where the system should not act alone.

Proves You can apply AI to a real workflow instead of forcing a model into the problem.

  • RAG
  • Agents
  • Evals
  • Human in the loop
Forward Deployed Engineering · Rollout, Adoption and ExpansionFlagship

A forward-deployed production launch

Deployed to a pilot group, instrumented, hardened, documented and measured against both technical reliability and user adoption.

Proves You can stay with a system from discovery through production and adoption.

  • Deployment
  • Observability
  • Adoption
  • Enterprise
Capstone · CapstoneFlagship

Capstone

A problem you chose, scoped and briefed before you started, then built, measured, documented and reviewed.

Proves The piece your portfolio leads with.

  • End to end
Shared Core · Programming Foundations

API-backed command line tool

Pulls from a public API, stores what it finds, and ships with a README someone else can follow without asking you anything.

Proves You can write software other people can run.

  • Python
  • APIs
  • Git
Shared Core · Statistics and ML Intuition

An honest data analysis

A claim about a real dataset, tested properly, stating plainly where the evidence runs out.

Proves You know when a result is real.

  • Pandas
  • Statistics
ML Engineering · Supervised Learning

Predictive model on a business problem

A model that beat a baseline you established first, with the reason it beat it written down.

Proves You can model something that matters, and defend it.

  • scikit-learn
  • XGBoost
ML Engineering · Computer Vision

Image classifier on data you collected

Trained on images you gathered yourself, with the failure cases documented rather than hidden.

Proves You can run the whole loop, including the messy start.

  • PyTorch
  • CNNs
  • Transfer learning
ML Engineering · MLOps

Deployed model with monitoring

Behind an endpoint, tracked, and watched for drift — the version that survives contact with production.

Proves You can ship, not just train.

  • MLOps
  • Deployment
AI Engineering · Building AI Applications

A complete AI application

Built from scratch and deployed, with accounts, a database and streaming responses.

Proves You can build a product, not a demo.

  • FastAPI
  • Next.js
  • Auth
AI Engineering · AI Evaluation

An evaluation harness

Golden cases for your own RAG system — including a regression it caught that you would otherwise have shipped.

Proves You can measure whether an AI system got better.

  • Evals
  • Testing
Forward Deployed Engineering · Discovery and Problem Framing

A deployment brief for a real workflow

Maps the current workflow, constraints, systems, users and success metric before a line of production code is written.

Proves You can turn an ambiguous customer problem into an engineering plan.

  • Discovery
  • Systems mapping
  • Scoping
Forward Deployed Engineering · Rapid Prototyping

A working customer-facing prototype

A thin end-to-end workflow tested with users and changed based on what they actually did rather than what everyone predicted.

Proves You can move from problem to usable software quickly without losing judgement.

  • Prototyping
  • AI-native dev
  • User testing
Forward Deployed Engineering · Integration Engineering

An enterprise integration service

Connects real systems through authenticated APIs and events, survives common failures, and exposes a workflow another service can depend on.

Proves You can make new software work inside an existing technical environment.

  • APIs
  • OAuth
  • Webhooks
  • MCP

Every one of these is shippable. Not a notebook you would have to explain — a link someone can open, with the problem it solves and what failed in the first version written underneath. That write-up is what gets read.

Apply

Begin the first module

Become AI native, it is the real deal today, and if it is not for you, you have lost nothing but learnt a new skill.

Apply