Free AI/ML learning roadmap

Artificial intelligence software learns patterns from examples, and machine learning engineers turn that into products people can actually use. Giuliano Liguori’s new roadmap organizes that path from school-level math to deployment tools used in production systems. (x.com) The guide starts with foundations: algebra, linear algebra, calculus, probability, and statistics. It points readers to free courses from Khan Academy and visual math lessons from 3Blue1Brown before moving into model-building code. (khanacademy.org) (3blue1brown.com) From there, the roadmap shifts into PyTorch, the open-source framework for building and training neural networks, the software pattern behind many modern image and language models. PyTorch describes itself as a framework for tensors and dynamic neural networks with strong graphics-processor acceleration. (pytorch.org) (github.com) The roadmap then moves past model training into Machine Learning Operations, or MLOps, the work of packaging, testing, deploying, and monitoring models after they leave a notebook. That production step is where many self-study guides thin out, even though startup teams often need one engineer to handle the full stack. (github.com) (geeksforgeeks.org) Its tool list reflects that emphasis. FastAPI is included for serving models through application programming interfaces, and Hugging Face documentation is included for using pretrained models, datasets, and inference tools. (fastapi.tiangolo.com) (huggingface.co) FastAPI’s own documentation says the framework is designed for Python APIs with standard type hints and high performance. Hugging Face’s Transformers docs say the library supports inference and training across text, vision, audio, video, and multimodal models, with more than 1 million model checkpoints on the Hub. (fastapi.tiangolo.com) (huggingface.co) That makes the roadmap less like a university syllabus and more like an engineering checklist. The sequence runs from math prerequisites to model code to web serving, which matches how small companies often build machine learning features: train something useful, wrap it in an API, and ship it. (x.com) (fastapi.tiangolo.com) The post also lands into a crowded market of AI learning maps, from roadmap.sh’s machine learning track to GitHub repositories promising six-month plans and interview prep. Liguori’s version stands out mainly by centering production work instead of stopping at theory or Kaggle-style experiments. (roadmap.sh) (github.com) For engineers trying to break into machine learning in 2026, the practical question is no longer just how to train a model. It is whether they can connect math, code, deployment, and maintenance into one working system, and that is the path this roadmap lays out. (x.com)