Pythonᴬᴵ: Build Your Own AI

Students train their own image classifier using Google’s Teachable Machine. They collect training data through their webcam, hit train, and watch their AI recognize objects in real time. They discover that when their training data is missing certain examples, their AI fails badly. Biased data makes biased AI. This sets the vision for the whole course.

Students learn dictionaries — Python’s way of organizing data with meaningful labels instead of numbered positions. Instead of remembering that item [0] is a name and item [1] is an age, they can just ask for data[“name”]. They build a knowledge base and discover this is exactly how real AI chatbots store the information they look up when answering questions.

Students learn to iterate through lists and dictionaries — processing every item automatically. This is how real programs handle data at scale: go through every student in a class, every song in a playlist, every pixel in an image. They combine this with AI-generated datasets to build something that feels like real data analysis.

Students learn to package their code into reusable functions — named blocks they can call whenever they need them. This makes programs organized, readable, and powerful. They also discover that this is exactly how AI agents work: when you ask an AI to “search the web” or “set a timer,” it’s calling a function — just like the ones students write in this lesson.

Students level up their functions to accept inputs and send back outputs. This is the input → process → output pattern that powers every app and AI service in the world. When you upload a photo and get back “this is a cat,” that’s a function: image goes in, answer comes out.

Students learn import — the key that unlocks Python’s massive ecosystem. They discover that professional developers don’t build everything from scratch; they use libraries that other people have already built and tested. Students explore built-in modules, install their first external package, and use a beginner-friendly data library to process a real dataset.

Students learn that real programs crash — and good programmers plan for it. They master try/except to handle errors gracefully instead of letting the program blow up. On the AI side, they discover prompt injection: how cleverly worded inputs can trick AI into ignoring its rules. They attack an AI prompt, then build defenses against it — thinking like both a hacker and a security engineer.

Students master Python’s string methods — the same operations that power every text-processing AI. They learn to slice, search, replace, and format text, then discover tokenization: the very first thing every AI language model does when it receives your message. AI doesn’t read words like you do — it breaks them into pieces called tokens.

The big reveal: machine learning is not magic. It’s just teaching a computer to find patterns in data — the same way students already think about if/else logic, but automated. Students learn what it means to train a model on examples, and use scikit-learn to build a decision tree: an AI that learns its own if/else rules from data instead of having a human write them.

Students apply everything from Lesson 9 to build something they can actually show off: a predictor that takes real input and makes a useful guess. They go through the full ML workflow — choosing a question, preparing data, training a model, testing it, and improving it. They discover that most of the work in ML isn’t the code — it’s choosing good data and asking the right question.

Students begin building a Smart Assistant — a Python application that combines every skill from the course. The assistant understands user commands, looks up information from a knowledge base, performs actions through functions, processes text intelligently, handles errors without crashing, and uses a trained ML model to make predictions. It’s the ultimate project: everything they’ve learned, in one application.

Students finish, polish, and present their Smart Assistants to the class. They add more commands, improve responses, expand the knowledge base, and add personality. The class then tries each other’s assistants live. The course closes with the full arc: in Lesson 1, they trained AI by clicking buttons. Now they’ve built an intelligent application from scratch.