Multimodal Annotation & Prompt Ranking for AI Training
Worked with Toloka on large-scale multimodal AI training projects involving text, image, and video annotation, and currently continue to contribute to ongoing tasks. For language model projects, performed prompt ranking and preference labeling, evaluating multiple prompts or AI-generated responses and categorizing the best-performing option based on clarity, relevance, instruction adherence, and guideline compliance. This work directly supports model fine-tuning and reinforcement learning workflows. For computer vision tasks, annotated images and videos by labeling objects, actions, and visual attributes to support object detection, classification, and action recognition models. Ensured high annotation accuracy at both the image and frame level while adhering to detailed task guidelines and quality benchmarks. Maintained consistent performance across high-volume workloads and delivered reliable, high-quality annotations under strict QA standards.