Aether
Worked on large-scale human-in-the-loop AI training projects involving multimodal data (text, images, and short-form video). The scope included evaluating and ranking AI-generated images against original prompts using predefined quality metrics such as prompt adherence, visual accuracy, realism, and entity consistency. Performed image annotation by localizing target entities with bounding boxes, and classified images and videos based on entity presence and relevance. Conducted entity recognition and linking by aligning structured knowledge sources (e.g., Wikipedia entries) with unstructured multimedia content. Authored and refined prompts for generative image editing, including entity removal and inverse prompts to reconstruct original images. Contributed to dataset curation by selecting and validating high-quality samples for supervised fine-tuning and evaluation. Adhered strictly to project guidelines, annotation standards, and multi-stage quality checks, ensuring consistency, accura