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

Multimodal: shared embedding space

From: Beyond the Vector →

Text and images embed into one shared space (CLIP-style), so a text query lands near the matching image and retrieves it directly, with no words on the image required.

Shared space: a text query points straight at an imageinteractive

Everything, words and pictures, is embedded into the same space by a CLIP-style model.

Multimodal: translate to text first

From: Beyond the Vector →

Turn each non-text item into text first, captioning images, parsing tables, transcribing audio, then retrieve over the descriptions with ordinary text RAG.

Translate first: describe the picture, then search the wordsinteractive

Turn each non-text item into text first: caption images, parse tables, transcribe audio. Then retrieve over the descriptions with ordinary text RAG.

Cache-augmented generation (CAG)

From: Beyond the Vector →

When the whole knowledge base fits in the context window, preload it once, compute the model's key-value cache over it, and reuse that cache for every query. No retrieval step, no vector database.

CAG: preload once, reuse the cacheinteractive

Preload the whole small corpus once and compute the model's key-value cache over it. This happens a single time.