At Perplexity AI, I validate benchmark instructions, source materials, reference solutions, and evaluation criteria. I inspect agent traces and audit grading logic, then document evidence-based findings and actionable improvements.
On Project Cursa (V2), I annotate robot-manipulation videos using synchronized camera views. I segment timelines and describe actions at atomic-movement, subtask, and task levels, recording observed outcomes accurately, including failed grasps and dropped objects.
At Alignerr, I transcribe and review Brazilian Portuguese audio, checking word-level accuracy, timestamps, speaker identification, and audio-text alignment. My other work includes evaluating AI responses, reviewing benchmark evidence, and contributing to open-source projects through GitHub.
