From zero evals to a working multimodal evaluation in 30 minutes using LangWatch Skills

The article describes building evaluations for an AI agent (InField Agent) that handles multimodal tasks including knowledge base retrieval, station status queries, and satellite image NDVI analysis. Using LangWatch Skills integrated via Claude Code, the author went from zero testing to a working multimodal evaluation pipeline in 30 minutes, covering tracing, scenario-based testing, and a path to production deployment on AWS.