Built an OSS toolkit to normalize mixed-script/mixed-language user text before LLM pipelines. Focus is production use: API mode, Docker deploy path, language-pack interface, and evaluation snapshots. Would appreciate feedback on data strategy, evaluation design, and integration patterns. https://github.com/SudhirGadhvi/open-vernacular-ai-kit