SEATauBench

Featured
AI/ML Research -
LLM Evaluation Agentic AI Southeast Asian Languages Localization SEACrowd

As part of the SEACrowd 2026 Research Apprenticeship, I extended Tau2-Bench, a Tool-Agent-User benchmark for real-world domains, into SEATauBench for low-resource Southeast Asian languages and localized task contexts.

I designed and ran a translation and localization pipeline across five Southeast Asian languages. The pipeline preserved executable layers while localizing agent-facing tasks, policies, databases, schemas, and tool surfaces.

I also analyzed multi-turn agent-user-tool trajectories to diagnose performance degradation and language-use failures.

Mentors: Dr. Samuel Cahyawijaya and Patomporn Payoungkhamdee.