AI Curriculum program for Tomorrow School
The picture of you, not of the topic. Each statement is something you'll be able to do by the end.
I can Develop and Deploy AI Solutions
I can Perform Data Analysis and Visualization
I can Engineer and Evaluate Machine Learning Models
I can Utilize Advanced AI Techniques and Tools
I can Implement Software Engineering Best Practices
The path through the material. Each lesson tackles one essential question.
How do Python’s core structures let us turn raw input into shaped, reliable data we can reason about, while controlling flow and failure cleanly?
How can we design transformations and models so small pieces compose into clear, reusable programs?
What turns a one-off script into a dependable command others can run on real data?
How do we structure code and history so we can change things boldly without breaking what already works?
What practices let us transform messy code into a fast, reliable library we can trust over time?
How does structuring raw files into arrays and DataFrames unlock fast, reliable computation while handling missing values and controlled randomness?
How can numerical linear algebra and probability distributions work together to describe data and support sound, quantitative predictions?
Why does decomposing matrices reveal the directions that capture variation and make linear models both interpretable and efficient to compute?
How do probabilistic models and hypothesis tests turn noisy data into defensible conclusions while quantifying uncertainty and tradeoffs?
What habits and table operations allow us to move from an open-ended question to a reproducible, defensible story supported by careful exploration?
How do thoughtful summaries and visuals turn raw tables into patterns we can reason with and test?
How do we shape messy data and exploratory visuals into a coherent argument that withstands scrutiny?
How does choosing the right representation—over time or text—unlock effective retrieval and decision-making?
What practices make our transformations and model calls predictable, debuggable, and cost-aware as datasets and systems grow?
How can we structure prompts and feedback loops so language models solve problems consistently and transparently?
How do we turn a probabilistic LLM into a dependable, typed component in a production system?
What practices turn an LLM-driven prototype into a measured, reproducible pipeline we can trust to run anywhere?
Why does thoughtful feature representation, packaged in a pipeline, unlock reliable learning and retrieval across datasets?
How does a probabilistic linear model translate scores into decisions, and what do our metric choices reveal about tradeoffs?
What makes a classifier trustworthy beyond the training set, not just accurate on yesterday's data?
How does gradient-driven learning in layered networks create reusable features for new tasks?
How do we shape optimization and model capacity so that performance improves on unseen data rather than just the training set?
How does embedding prior knowledge into our models and data pipeline help them generalize when examples are scarce?
How can we make model choices we trust by combining diverse predictors and rigorous, reproducible evidence?
What closes the gap between fluent answers and truthful, source-grounded responses when language models consult a corpus?
What practices keep an ML service honest and reproducible from experiment to containerized deployment?
Three ways to connect: Claude Code (PAT + install command), Claude Desktop (.mcpb download — no token to paste), or Claude web (Customise → Connectors → Add custom connector, OAuth). Same MCP endpoint, same identity on every path.
https://nebular.live/api/v1/mcp/