At OneForma, I reviewed AI-generated responses for accuracy, relevance, safety, reasoning, and instruction compliance. I used project-specific scoring frameworks and wrote structured feedback to support model improvement.
I analyzed challenging outputs for hallucinations, contradictions, weak reasoning, bias, and edge-case failures. I also compared model responses through preference-ranking tasks supporting human feedback and reinforcement learning.
With TELUS International AI / UHRS / Toloka Programs, I assessed search relevance and ranking, classified online content, and contributed to A/B testing. A structured review and edge-case checking process I introduced helped lower review rejections by 20%.
At Scale AI, I annotated audio for speech recognition, speaker diarization, and intent classification, and produced computer vision labels. I maintained 98%+ accuracy on gold-standard tasks and helped reduce downstream QA rejections by 25% through validation and careful review.
