JSON for the payload, XML tags for the wrapper. Which format survives nesting, partial streams, and a model that wandered off-schema.
JSON and XML tags serve distinct roles: JSON defines schema, XML delimiters mark output location; combining both (`<result>{json}</result>`) handles model preambles. For tech leads: format choice affects reliability and cost; raw JSON with constrained decoding, XML for shallow multi-section output reduces complexity.
Book: Prompt Engineering Pocket Guide: Techniques for Getting the Most from LLMs
Also by me: Thinking in Go (2-book series) — Complete Guide to Go Programming + Hexagonal Architecture in Go
My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools
Me: xgabriel.com | GitHub
Your extraction endpoint has run clean for weeks. Then a support ticket arrives in mixed German and English, the model decides to be helpful, and the response comes back as:
A 1.5B model running on your laptop will return JSON that almost parses. The closing brace is...
Thoughtworks Technology Radar Vol 34 (April 2026) puts Structured output from LLMs in the Adopt...
A benchmark of how OpenAI, Anthropic, and Google Gemini break under realistic JSON schemas, and how to engineer around it.
Extractive vs abstractive compression of retrieved chunks. Sentence-level filtering. How to cut tokens without losing the answer.
I've been building an extraction API for the past month. The use case is specific — reading informal...
If you stream tool calls or structured output from an LLM, you have almost certainly seen one of...
