All Articles

Practical lessons from building production RAG systems, covering chunking strategies, hybrid search, reranking, and evaluation.

A comprehensive look at Anthropic's Claude 3.5 Sonnet, covering benchmarks, real-world performance, and practical applications.

An AI researcher documents replacing her production ML pipeline with open-source alternatives. Inference costs dropped 94% while quality decreased only 3%. Includes exact benchmarks, cost tables, and honest assessment of tradeoffs.

A visual guide to understanding attention mechanisms in transformers, with intuitive explanations and minimal math.