Agentic coding in the terminal. This site was built with it.

Type-safe React framework with SSR and server routes. Powers this site.

One API for many LLM providers, so swapping models is a config change, not a rewrite. Backs the chatbot here.

Python package and project manager that replaces pip, venv, and pyenv, and is fast enough that you stop noticing installs.

Linter and formatter for Python in one binary. Replaces flake8, isort, and black.

Default place to find pretrained models and datasets before training anything from scratch.

Free desktop GIS. The quickest way to eyeball rasters and vector labels before they go into a model.

Petabytes of satellite imagery with compute next to it. Good for building remote-sensing datasets without downloading scenes.

Hand-drawn style whiteboard for sketching architecture diagrams fast.

The clearest guide on when to use a simple workflow and when an agent actually earns its complexity.

Deep, well-cited surveys on LLM agents, prompting, and hallucination. Read before a new topic.

Daily, hands-on notes on new LLM releases and tools. The fastest way to keep up without the hype.

Applied ML and LLM system patterns from production: evals, RecSys, and guardrails.

ML engineering and MLOps from someone who wrote the book on it.

The visual explanation of attention to send anyone starting with transformers.

Builds backprop, then a GPT, from scratch in code. The best way to really understand what training does.

Visual intuition for gradient descent, backprop, and attention. Pairs well with the Karpathy series.

Top-down and code-first: you train working models in lesson one, then learn why they work.

Free, hands-on course on transformers, fine-tuning, and the Hugging Face stack.

One-hour courses on specific GenAI topics (RAG, agents, evals). Good for filling one gap quickly.

The transformer paper. Short enough to read in one sitting.

Where the term RAG comes from. Useful context before building any retrieval system.

The reason-then-act loop that most tool-using agents still follow.

Promptable segmentation that changed how quickly you can label images for CV and remote sensing.