At Hugo Technologies, I design adversarial red-teaming tasks to stress-test frontier LLM and CUA research capabilities, exposing logic traps and search loop vulnerabilities.
I audit multi-step web search trajectories and use OSINT research across academic archives, public records, and primary sources to build benchmark prompts with unambiguous, primary-sourced ground truths.
I evaluate responses for safety, policy compliance, and factual accuracy, then analyze failure patterns to document model limitations and provide actionable guidelines for engineering and alignment teams. Previously, as a Data Verification Specialist, I cross-checked records against primary sources and authored documentation standards and quality evaluation rubrics.
