United States
Annotate and evaluate responses from large language models (LLMs) including GPT-4, Claude, and Gemini for accuracy, relevance, safety, and factual correctness across diverse domains. Perform RLHF tasks, ranking model outputs and providing detailed preference-based feedback to align model behavior with human values and intentions. Execute complex coding, reasoning, mathematical, and creative writing prompts to assess model capabilities and identify failure modes. Maintain 97%+ accuracy rating across 2,500+ annotation tasks with consistent quality scores in the top 5% of contractors on the platform. Collaborate with ML engineers to identify edge cases, refine annotation guidelines, and improve evaluation rubrics for specialized domains including STEM and legal reasoning. Conduct red-teaming exercises to surface potential safety risks, biases, and adversarial vulnerabilities in model outputs.
Completed diverse generalist AI evaluation tasks including text classification, image captioning validation, fact-checking, and comparative model ranking. Demonstrated versatility across multiple project types, rapidly adapting to new instructions, rubrics, and evaluation criteria with minimal onboarding time. Built a reputation for reliability and speed, achieving 'Top Contributor' status on multiple crowdsourced AI training platforms. Developed personal workflows and checklists to maintain accuracy while maximizing task throughput in fast-paced, quota-driven environments.
Created precise bounding boxes, polygons, and semantic segmentation masks for autonomous vehicle training datasets using Labelbox and CVAT platforms. Annotated 10,000+ image frames with 99.2% inter-annotator agreement, exceeding platform quality benchmarks by 8%. Applied rigorous QA protocols to identify and correct inconsistent labels, reducing dataset error rates by 18% through systematic review processes. Processed high-volume annotation pipelines under tight deadlines, consistently meeting or exceeding daily throughput targets by 15%.
Labeled and categorized user-generated content for sentiment analysis and content moderation AI systems across multiple languages and cultural contexts. Evaluated search engine results and ad relevance for major tech clients, improving ranking algorithm accuracy through detailed relevance scoring. Completed transcription and audio annotation tasks for speech recognition model training with 98% accuracy and high consistency ratings. Adapted quickly to evolving project guidelines and platform updates, maintaining productivity during high-demand periods.