Library
These are the books, courses and tools I study with and keep coming back to. I've worked through every one of them and found them incredibly helpful. If you want to understand how modern AI works, these are good places to start.
Theory and Practice of Neural Networks, Computer Vision, Natural Language Processing, and Transformers Using TensorFlow
A full-colour tour from a single artificial neuron to fully connected, convolutional and recurrent networks, and on to the Transformer, GPT and BERT. Each idea comes with code, in TensorFlow in the book and in PyTorch in the companion repository.
Builds a GPT-style model step by step on an ordinary laptop: tokenization, attention, the transformer block, pretraining, then fine-tuning for classification and for following instructions.
Picks up where a working LLM leaves off and adds reasoning: evaluating answers, inference-time techniques that spend more compute per question, and training methods such as reinforcement learning and distillation.
Efficient AI for local deployment
How to build, adapt and optimize small, focused language models for one domain, and deploy them on commodity hardware instead of relying on a giant general-purpose model.
A complete technical computing environment: symbolic and numeric maths, plotting, data and machine learning in one notebook, with curated Wolfram|Alpha knowledge built in. It is a quick way to check the maths behind the books above, such as a matrix product, a softmax, or a gradient, exactly and numerically, and to plot it.
The plan I use
Current Personal pricing ↗ Student plan ↗ Wolfram Account ↗
From Wolfram's pricing page, September 2026. Prices and terms change; check before you buy.
Andrew Ng's AI education platform: full specializations, dozens of free short courses on LLMs, agents and fine-tuning, often taught with the teams that build the tools, and The Batch, a weekly newsletter on what's happening in AI.