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.

Portrait of Sachin Gupta rendered in binary

Swap the model, keep the harness

From: The Harness Is the Agent →

Pick a different model and watch the six harness responsibilities stay identical. The determinism, tests, and governance live in the harness, not the model.

Swap the model:

model (swappable)

Claude

harness (unchanged)

Control loop — when to call the model, when to stop

Tool interface — what actions exist and how they are validated

Context assembly — what goes into the prompt each turn

State & memory — what carries across steps

Guardrails — what the agent is allowed to do

Verification — is the output grounded and safe

Whichever model you pick, the six responsibilities do not move. The determinism, the tests, and the governance live in the harness, not the model. That is why the harness is the agent.

Cosine similarity & distance

From: How RAG Finds Things →

Two pieces of text become two vectors. Cosine measures the angle between them, not their length. Drag the arrowheads and watch the score move.

Drag either handle · A = solid · B = dashed

Cosine similarity

1.00

-101

Angle: 0°
Dot A·B: 11.9

Nearly the same direction → treated as the same meaning.

cos(θ) = (A·B) / (|A|×|B|)

The system design routine (six steps)

From: How to Design a System →

How a vague 'design X' prompt becomes a real design. Watch the six-step routine, scope, estimate, API, data model, high-level design, then deep dive, walk itself across a link shortener, so you never freeze on a blank page.

A routine for any "design X" questionauto-playing

Someone says 'design a link shortener.' A blank page. The urge is to freeze, or to jump straight to a database with no plan.