Handshake AI
Image
Bounding BoxI contributed to Project Hedgehog as a multimodal evaluator working across image, video, text, and audio tasks. My responsibilities included identifying AI‑generation artifacts, assessing visual quality, verifying reference preservation, evaluating instruction following, and performing detailed side‑by‑side comparisons of model outputs. I worked extensively with complex image transformations, video inpainting and object tracking, audio transcription accuracy, and text‑based instruction compliance. This required precise attention to detail, consistent rubric‑based judgment, and the ability to spot subtle inconsistencies across different media types. Throughout the project, I handled a wide range of evaluation formats, including R2I comparisons, video segmentation audits, Instagram entity tagging, audio WER assessments, and multimodal instruction‑following checks. This breadth of experience allowed me to develop a strong understanding of model behavior across modalities and to provide high‑quality, reliable annotations that directly improved model performance. My work consistently met or exceeded quality benchmarks, and I became highly proficient at identifying error patterns and ensuring data integrity across all task types.