Learning with AI
Open, searchable study guides and interactive books I build and maintain on modern AI — from ML foundations to building, evaluating, and serving agents and LLMs in production.
A growing collection of open study guides I maintain. Each one is a full, searchable book you can read online. Pick a track below.
Machine Learning & LLMs
Classical ML, transformers, attention, LLM inference, alignment/RLHF, and scaling laws.
Open guide → Production Agent Engineering
Build-first: tool design and MCP, context engineering, memory, orchestration, CI/CD, security, deployment.
Open guide → Agentic AI Evaluation
Metrics, benchmarks, tool-use and safety evaluation, monitoring, plus a design-patterns playbook.
Open guide → LLM Inference & Serving
vLLM, Triton, Kubernetes, autoscaling, canary deploys, monitoring, and drift detection.
Open guide → Toxicogenomics & Single-Cell FMs
ML for toxicogenomics and single-cell foundation models: biological response and cell-state representations.
Open guide → Elsewhere on the web
Guides and books by other people that I keep coming back to. All free to read online.
Courses
- RLHF & Post-Training Course — Nathan Lambert’s course on RLHF and post-training, companion to his RLHF book. Slides and notes here, lecture videos on YouTube.
- Cohere Labs ML Summer School — recorded lecture series from Cohere Labs’ Open Science Community, spanning ML maths through distributed training. Programme page.
- Hugging Face LLM Course — practical, hands-on, and kept current with the ecosystem.
- Hugging Face Agents Course — agent fundamentals through deployment, with hands-on units.
- Full Stack Deep Learning — everything around the model: data, testing, deployment, monitoring, teams.
Foundations
- Dive into Deep Learning — an interactive book where every concept comes with runnable code in PyTorch, TensorFlow, and JAX.
- Understanding Deep Learning — Simon Prince’s book, free as a PDF, with unusually clear figures.
- Mathematics for Machine Learning — the linear algebra, calculus, and probability you actually need.
- The Little Book of Deep Learning — the field compressed into an evening’s reading. Good for revision.
LLM internals
- Neural Networks: Zero to Hero — Karpathy builds backprop, then a transformer, then a tokenizer, from nothing.
- The Ultra-Scale Playbook — how training actually works across thousands of GPUs.
Agents
- Building Effective Agents — when to use an agent and when a workflow is the better call.
- Model Context Protocol — the spec and docs for the emerging standard for connecting agents to tools.
Systems and serving
- CUDA Kernels: GPU & Parallel Programming from First Principles — GPU programming built up from fundamentals.
- ML Engineering Open Book — field notes from training large models: hardware, debugging, failure modes at scale.
Practice
- Made With ML — design, develop, deploy, and iterate on production ML systems, with code.