Independent researcher and practitioner working on AI-agent evaluation, execution trajectories, delegated authority, control integrity, and practical agentic systems.
Nirmata studies what final-output evaluation misses when AI agents use tools, cross authorization boundaries, change state, and create external effects.
Current directions:
- trajectory-aware evaluation of AI agents;
- human approval and delegated-authority protocols;
- factorized content/process evaluators;
- failure-preserving and reproducible agent evaluation;
- applied cases in publishing, browser automation, and business operations.
Research materials are published in English and Brazilian Portuguese.
Pesquisadora independente e profissional dedicada à avaliação de agentes de IA, trajetórias de execução, autoridade delegada, integridade de controle e sistemas agênticos aplicados.
O Nirmata investiga o que a avaliação apenas do resultado final deixa de perceber quando agentes usam ferramentas, atravessam limites de autorização, alteram estados e produzem efeitos externos.