Examples with LLMs demonstrate how data without metadata, like temperature units, leads to inconsistent and unreliable interpretations.
Providing a schema with explicit constraints and descriptions, such as units, enables LLMs to interpret data correctly and generate consistent results.
Data schemas should be created first and be independent of code assets, allowing code generation for polyglot applications and improving LLM performance.
Using a standard like CloudEvents provides a consistent envelope for events, preventing the chaotic and varied structures that LLMs might otherwise generate.
By providing a comprehensive metadata document that includes endpoint and protocol details, LLMs can automatically generate complete and correct client code for publishers.
Current event pipelines often rely on informal communication, but a metadata-driven approach creates a formal contract that improves maintainability and compatibility.
A real-world example processing USGS water data demonstrates how a complete metadata definition using X Registry enables automated tooling and contract enforcement.
JSON Structure is proposed as a stricter alternative to JSON Schema, designed as a data definition language with a better type system and namespace support.
A summary of three key open standards—CloudEvents, X Registry, and JSON Structure—that provide the tools for building robust, metadata-driven systems.
20:29Clemens Vasters
21:17Lars Kölker
38:36Ferd Scheepers
45:36Clemens Vasters
19:44Dominik Schneider
29:25Matthias Niehoff
32:03Eldert Grootenboer & Olena Borzenko-Turianska
30:51Soroosh Khodami
Jobs that call for the skills explored in this talk.
AUTO1 Group SE
Berlin, Germany
Thinkport GmbH