From your first prompt to a production AI system — in 16 weeks

Week01Days 1–7

You shipYour first LLM call from your own environment

AI Foundations: Python for AI Engineers

The map of AI on day one, then the Python every model and agent will stand on.

The AI landscape: ML vs DL vs NLP vs GenAIThe stack: model → prompt → context → harness → loop → graphPython I–IV: syntax, control flow, functions, OOPPython V: generators, iterators, decoratorsEnvironment setup: Python, git & API keys

Stack: Python · git · API keys · your first LLM call from code

?Checkpoint ? The week closes with your environment fully working: Python written across five sessions, git and API keys configured, and your first LLM call made from your own code. You leave knowing exactly where ML, DL, NLP and GenAI sit in the stack.

Week02Days 8–14

You shipYour first trained and evaluated ML model

Machine Learning Fundamentals

The discipline underneath everything: teach a machine to learn from data.

Supervised vs unsupervised vs reinforcementRegression & classification, hands onTrain/test splits, overfitting & regularizationEvaluation: accuracy, precision, recallClustering & dimensionality reduction

Stack: scikit-learn · pandas · NumPy · an ML mini-project end to end

?Checkpoint ? You now understand how machines learn from data: framing a problem, training regression and classification models, splitting data honestly, and reading accuracy, precision and recall. Your evaluated mini-project proves you can run the full ML workflow end to end.

Week03Days 15–21

You shipA neural text classifier, trained and tested by you

Deep Learning & NLP Fundamentals

From neurons to language: the road that leads straight to the transformer.

Neural networks & backpropagation intuitionA taste of PyTorch: build a small networkTokenization & embeddings: vectors of meaningNLP tasks: classification, NER, summarizationFrom RNNs to attention: why transformers won

Stack: PyTorch · Hugging Face · embeddings · NLP pipelines

?Checkpoint ? Neural networks are no longer a black box: you have built one, trained it in PyTorch, and turned text into embeddings. With tokenization, classic NLP tasks and the road from RNNs to attention behind you, the transformer is ready to be understood.

Week04Days 22–28

You shipA production-grade system prompt with its own evaluation

Prompt Engineering

Shape a single call to the model, then hit the prompt's ceiling deliberately.

System prompt engineeringPrompt anatomy: zero / one / few-shotChain-of-Thought & reasoning-time techniquesEvaluating, optimising & attacking promptsWhy the prompt alone is not enough

Stack: Frontier chat models · API playgrounds · prompt evals

?Checkpoint ? You can now engineer a prompt the way production teams do: structured system prompts, shot-based and reasoning techniques, and a proper evaluation loop that catches failures. You have also felt the prompt's ceiling, which is exactly why the next weeks exist.