AI Project Contributor public Remote experience
Handshake AI
May 2026 - Current (6 months)
Designed 13 adversarial test cases targeting visual reasoning failures in multimodal AI models, focusing on structured data such as charts, tables, and diagrams.
Successfully induced 8 verified model failures per accepted task, maintaining a 100% acceptance rate with zero rejected submissions.
Identified failure mode patterns among 5 different failure reasons, informing evaluation strategy for black box vs. transparent model outputs.