Learn

Things I've worked with, written down.

Structured series — read in order, or jump to what you need.

Machine Learning

Machine Learning from First Principles

Machine Learning

30 parts · 5 published

From a line through five houses to large language models. Every topic goes idea → math → code → practice, with derivations to do on a whiteboard and interview questions stacked from phone screen to research level.

Start reading →
✓01Linear Regression: Drawing the Best LineDerivationsInterview Bank
✓02Optimization: How Models Actually LearnDerivationsInterview Bank
✓03Logistic Regression and ClassificationDerivationsInterview Bank
✓04Generalization: Overfitting, Bias–Variance and RegularizationDerivationsInterview Bank
✓05Evaluation: Measuring Models HonestlyDerivationsInterview Bank
○06Probability, Likelihood and Bayes
○07Causal Inference for Data Scientists
○08Generative Classifiers and kNN
○09Decision Trees
○10Ensembles: Bagging, Random Forests and Boosting
○11SVMs and Kernels
○12Clustering
○13Dimensionality Reduction
○14Time Series and Forecasting
○15Recommender Systems
○16Neural Networks and Backpropagation
○17Training Deep Networks
○18Convolutional Networks and Vision
○19Sequence Models
○20Representation Learning
○21Attention and the Transformer
○22Language Modeling and Pretraining
○23Scaling
○24Post-training
○25Inference and Efficiency
○26LLM Systems
○27ML System Design
○28MLOps
○29Responsible ML
○30How It All Comes Together

LLMs & Agents

LLMs from Zero to Hero

LLM Engineering

17 parts · 17 published

A complete guide to understanding, building, and deploying large language models — from tokenization and attention to fine-tuning, alignment, and the frontier. Written for practitioners who want to go deep.

Start reading →
✓01A Brief History of Language Models→✓02Tokens, Embeddings, and Attention→
✓03Attention: A Deep Dive with IntuitionDerivationsInterview Bank
✓04From GPT to LLaMA: Modern LLM Architectures→✓05Architecture Improvements That Actually Matter→✓06Pre-training Data and Tokenization at Scale→✓07Scaling Laws: How Models Get Better→✓08Supervised Fine-tuning: From Model to Assistant→✓09LoRA and QLoRA: Fine-tuning on a Budget→✓10RLHF and DPO: Alignment from Human Feedback→✓11Quantization: Smaller, Faster, Still Good→✓12Inference Engines and Serving LLMs in Production→✓13Retrieval-Augmented Generation (RAG)→✓14Mixture of Experts: Scaling Without the Compute→✓15State Space Models and Mamba→✓16What's Next: The Frontier of LLM Research→✓17LLM Interview Cheatsheet→

Building AI Agents That Work

AI Engineering

6 parts · 6 published

From the agent loop to multi-agent orchestration — how to build agents that survive production, not just demos. Covers tool use, memory, MCP, A2A, and the hardest part: knowing if your agent actually works.

Start reading →
✓01How AI Agents Actually Work→✓02Why Your Agent Works in the Demo and Fails at Work→✓03Agent Memory: How Agents Remember Across Conversations→✓04MCP: The Protocol That Standardized How Agents Use Tools→✓05Multi-Agent Systems: Orchestration Patterns That Work→✓06Evaluating Agents: How to Know If Your Agent Actually Works→

LLMOps

LLMOps: Running AI Systems in Production

AI Engineering

4 parts · 4 published

The operational side of LLMs that nobody talks about until something breaks — monitoring, cost optimization, model routing, and managing prompts like production code.

Start reading →
✓01Monitoring LLMs in Production: What Actually Matters→✓02The Economics of Running LLMs: How to Cut Costs Without Killing Quality→✓03Model Routing: Sending the Right Query to the Right Model→✓04Prompt Versioning: Managing Prompts Like Production Code→

Stay in the loop

New pieces, straight to your inbox.

Causal inference, experimentation, GenAI. No noise — just the next piece when it's ready.