Visualizers
Learn by moving things
Interactive explainers for the ideas behind retrieval, embeddings, and agents. A formula tells you what is true; dragging a vector shows you why. Each one also appears inside the article it belongs to, and runs entirely in your browser, so drag, poke, and break it.
How MCP evolved
From: How MCP Grew →An interactive timeline of MCP's four spec revisions. Click each to see what it added, from the initial primitives to authorization, async tasks, and governance.
The initial release. JSON-RPC base, capability negotiation, five primitives (tools, resources, prompts, sampling, roots), and the stdio and original HTTP+SSE transports.
Which grader catches what
From: How to Actually Evaluate an LLM System →Evaluating an LLM is choosing the right grader. Click each family, exact checks, reference metrics, LLM-as-judge, human review, to see how it works, what it costs, and what it can and cannot catch.
- How it works
- string equals, a regex, a JSON-schema check
- Cost
- free, instant, deterministic
- Catches
- exact values and formats
- Misses
- anything open-ended or reworded
The offline/online eval flywheel
From: You Cannot Improve What You Cannot Measure →Step through how a production failure becomes a permanent test: offline eval gates the ship, online eval catches a surprise, and that surprise is captured back into the offline set so it never recurs silently.