Hillary Osei

Labeled and annotated audio and video content for Apple’s voice assistant, supporting AI model training and QA pipelines. Evaluated language data for accuracy, intent alignment, tone, and regional/cultural consistency to enhance NLP model performance. Conducted linguistic QA on AI-generated text and search responses to validate semantic relevance and context accuracy. Flagged problematic content to support Responsible AI and human-in-the-loop evaluation processes. Applied structured annotation guidelines and internal tools to ensure consistent data labeling and high annotation throughput.

Evaluate outputs from Large Language Models (LLMs) for relevance, factual accuracy, and alignment with user intent. Conduct search relevance assessments by reviewing AI-generated responses and determining their effectiveness in answering complex user queries. Provide ground truth annotations to improve LLM performance across various domains. Identify biased, misleading, or potentially harmful AI outputs in alignment with client guidelines. Work with proprietary tools to label data, assess model quality, and contribute to AI system improvement via human-in-the-loop workflows.

Enforced content moderation and platform policies for a major social media networking platform, ensuring ads and user-generated content complied with community standards. Reviewed flagged content for violations involving misinformation, hate speech, explicit material, and borderline policy edge cases. Evaluated advertising content for clarity, quality, and compliance with internal platform rules and regulatory guidance. Used internal policy guidelines and investigative research techniques to assess and categorize sensitive or ambiguous materials. Contributed to Trust & Safety efforts by identifying high-risk content and providing detailed rationale for enforcement decisions. Collaborated with moderation teams to enhance platform safety, reduce harm, and support inclusive user experiences.