A practical roadmap for learning Generative AI, from Python to AI agents

Generative AI moves fast, and it is easy to jump straight to the newest framework and get lost. A better approach is to build in layers, where each stage makes the next one easier. This roadmap follows the same order as the free tutorials on this site and the curriculum of the Applied Generative AI Mastery course.

Stage 1: Python foundations

Almost every Generative AI library is written for Python. You do not need to be an expert, but you should be comfortable with:

• Variables, data types and f-strings
• Lists, dictionaries and loops
• Functions, and later classes
• Installing packages with pip and reading and writing files

Start with Python variables and data types, then lists, dictionaries and comprehensions.

Stage 2: working with data

LLM applications are data applications. You will load documents, clean text, and measure results. Learn enough pandas to load and summarise a table, and enough statistics to read an average and a spread without being misled.

Read Analyse your first dataset with pandas and Descriptive statistics in Python.

Stage 3: core machine learning ideas

You do not need to train neural networks to build with LLMs, but a few machine learning concepts come up constantly: training versus test data, overfitting, evaluation metrics and vectors. Build your first machine learning model covers the essentials.

Stage 4: using LLMs directly

Now work with the models themselves. Run an open model locally so you can experiment for free, then learn how prompts change the output.

• Run open-source LLMs on your own computer with Ollama
• Prompt engineering basics: personas, few-shot and chain of thought
• Compare models using public leaderboards and benchmarks, and test them on your own task

Stage 5: building applications

This is where frameworks earn their place. Hugging Face gives you access to thousands of open models and datasets. LangChain connects models to tools, memory and data sources, and LangSmith traces what your application is doing. Retrieval-augmented generation (RAG) lets a model answer from your own documents.

Start with RAG from scratch in Python, which builds every step by hand so the frameworks make sense when you meet them.

Stage 6: advanced topics

• Agentic AI: systems that plan, call tools and loop until a task is done, built with LangGraph, CrewAI or AutoGen.
• Deployment and LLMOps: serving, monitoring, evaluating and safeguarding applications.
• Fine-tuning: adapting a model to your domain with LoRA and QLoRA.
• Multimodal models: generating and understanding images, audio and video.

These stages are taught live, with ten real-world projects, in Applied Generative AI Mastery.

How to make the roadmap stick

1. Build something at every stage. A script, a notebook, a small app. Reading alone fades quickly.
2. Keep projects small and finished. A working document chatbot beats an unfinished “AI platform”.
3. Read the official documentation. Libraries change quickly, and the docs are always the most current source. Our resources page links to them.
4. Learn with others. A live class or study group keeps you accountable. The free workshop is a good place to start.