Abdur-Rahman Butler, a machine learning engineer at Isaacus, released an interactive guide teaching how vector embeddings work, inspired by Brendan Bycroft's LLM visualization guide. It covers vectorisation, cosine similarity, semantic search, hierarchical retrieval, clustering, zero-shot classification, and deduplication, and includes a Wordle/GeoGuessr-inspired game called SemanticGeussr. It matters to practitioners as a hands-on educational resource for embedding concepts that existing resources cover only narrowly.
Two interactive visualizations of GPT-2-small's 32,070 token embeddings: a 2D t-SNE map with minimum spanning tree edges, and a hyperbolic (Poincaré ball) tree. Users can tap, zoom, and search tokens to explore nearest-neighbor relationships. No forward pass or context is used; only the raw embedding table is visualized.
Two interactive visualizations of GPT-2-small's 32,070 token embeddings: a 2D t-SNE map with minimum spanning tree edges, and a hyperbolic (Poincaré ball) tree. Users can tap, zoom, and search tokens to explore nearest-neighbor relationships. No forward pass or context is used; only the raw embedding table is visualized.
A Medium article published on July 20, 2026, explains the internal workings of embedding models, covering how text is converted into vector representations. The post targets developers and researchers seeking a deeper understanding of embedding model architecture.
A Medium article published on July 20, 2026, explains the internal workings of embedding models, covering how text is converted into vector representations. The post targets developers and researchers seeking a deeper understanding of embedding model architecture.
This tool builds visual knowledge graphs from local documents using Ollama for embeddings and graph visualization. It helps users explore relationships between concepts in their own data without relying on external APIs.
This tool builds visual knowledge graphs from local documents using Ollama for embeddings and graph visualization. It helps users explore relationships between concepts in their own data without relying on external APIs.
Semantra is a JavaScript library that performs semantic search entirely in the browser, using Snowflake's Arctic-Embed-S model to embed text chunks and cosine similarity ranking via WebGPU. It solves the problem of adding semantic search to web apps without server infrastructure, targeting developers who want client-side search capabilities.
Semantra is a JavaScript library that performs semantic search entirely in the browser, using Snowflake's Arctic-Embed-S model to embed text chunks and cosine similarity ranking via WebGPU. It solves the problem of adding semantic search to web apps without server infrastructure, targeting developers who want client-side search capabilities.
A Medium article explains rotary position embeddings (RoPE) through hand calculations, building on a previous piece about sinusoidal absolute positional encoding. It aims to demystify how RoPE encodes position information in transformer models.
A Medium article explains rotary position embeddings (RoPE) through hand calculations, building on a previous piece about sinusoidal absolute positional encoding. It aims to demystify how RoPE encodes position information in transformer models.