Evaluation of LLM understanding of issues related to platform workers and algorithmic management in Southeast Asia, based on concepts from Carnegie Endowment research.

Comparing human judgements vs LLM judges on Claude Opus 4.5 and Sonnet 4.5 responses to legal and agriculture questions in 7 Indian languages.

Human ratings from Karya platform (native speaker evaluators) are stored separately for comparison with LLM judge scores.

Prompts: 9293 Languages: Hindi, Bengali, Telugu, Kannada, Malayalam, Assamese, Marathi Domains: Legal, Agriculture Human evaluators: ~120 native speakers