OCDevel AI Podcast

Education

Learn AI and machine learning from the ground up - a complete, self-driving course that goes from "what is AI?" all the way to building and operating production AI systems. Every episode pairs a five-minute brief on the latest in AI with a tutorial that climbs a single ladder across ~100 episodes - interleaving the concepts, the math that actually explains them, hands-on code you build yourself, and the MLOps to ship it. It leaves no stone unturned: the probability, statistics, and Bayesian foundations most courses skip get the deep treatment they deserve, right alongside the practical work. The path runs from your first model on real data, through the mathematical bedrock, classical ML, neural networks built from scratch in PyTorch, transformers part by part, building with LLMs (RAG, fine-tuning, agents), and MLOps on AWS and GCP - to the capstone: operating a self-managing fleet of AI agents in production. The goal isn't a diploma, it's a job. Every phase leaves you a portfolio project, and the whole course is built to make you the rare "operator" who can ship real systems - the one-person AI department. For programmers who want to break into AI through self-directed learning - no grad school required. AI-generated podcast by OCDevel.

RSS feed

Classification Metrics That Matter: Confusion Matrix, Precision, Recall, F1, ROC/AUC, PR Curves, and Calibration

A classifier is two things at once: a probability ranker and a threshold, and almost every evaluation mistake comes from confusing the two. This episode is the model-agnostic ruler for measuring any classifier, from the confusion matrix up through ROC/AUC, PR curves, and calibration.

Logistic Regression and What Classification Really Is: Sigmoid, Log-Odds, Cross-Entropy, the Linear Decision Boundary, and Building One Neuron from Scratch

Logistic regression takes the same linear score from last episode, w dot x plus b, squashes it through a sigmoid into a probability, and trains it with cross-entropy, which turns out to be maximum likelihood for a Bernoulli label. It's a linear classifier and, quietly, a single neuron, the exact template every classification network stacks at its top.

Linear Regression: Your First Real Model, From a scikit-learn One-Liner to the Intuition Underneath

Fit a line, read its coefficients, and understand what "fitting" actually does before we ever touch the calculus. This is the dot product that quietly becomes logistic regression, a neuron, and every neural network downstream.

The Machine Learning Workflow and the Train/Test Split: Data, Features, Training, and Honest Evaluation

The whole point of machine learning is generalization to unseen data, and the single habit that protects it is holding out a test set you never touch until the end. We walk the full workflow loop from problem framing to monitoring, and we make the [train/test split](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html) and cross-validation concrete with scikit-learn.

NumPy and pandas: The Working Tools of Machine Learning

Before you touch a model, you need fluency in arrays and dataframes. This episode covers vectorization, broadcasting, views versus copies, axis semantics, and the pandas selection and groupby muscle memory every later ML episode silently assumes.

Setting Up Your AI Toolkit: Python, NumPy, Notebooks, Git, and Why a Portfolio Beats a Credential (Phase 0, Ep 4)

Build the exact dev environment this course runs on (Python, virtual environments, notebooks, git, VS Code) and learn why three to five shipped, defensible GitHub projects open more doors than any stack of certificates.

The Four Learning Paradigms: Supervised, Unsupervised, Reinforcement, and Self-Supervised

Every machine learning model fits one of four boxes, defined by where the training signal comes from: a human answer key, no key at all, a delayed reward, or labels the data manufactures from itself. Get this taxonomy crisp and you can place any model in Phase 1 and ace the most common ML interview opener.

A History of AI, Part 2: From the Dartmouth Workshop Through the AI Winters to Transformers and Agents (Phase 0, Episode 2)

Seventy years of AI swing like a pendulum between hand-coded logic and learning from data, and every winter for one camp was the opening for the other. This episode traces that arc from myth and the Dartmouth workshop, through two funding collapses, to the transformer and today's agents.

What Is AI? Untangling AI, Machine Learning, Deep Learning, Data Science, and Statistics (Phase 0, Episode 1)

AI is the goal, machine learning is the method, and deep learning is one branch, while statistics and data science overlap rather than nest inside. Get the mental map, the four learning paradigms, and the operator-versus-user thesis that the whole course is built on.