Rashi Garg

I’m an AI/LLM Trainer and Data Annotator who builds reliable model behavior through rigorous prompt writing, evaluation, and dataset work. I test how generative AI follows difficult instructions, then use findings to help improve helpfulness and accuracy for end-users.

To reduce risk, I compare AI-generated content against trusted sources and fix hallucinations using hallucination mitigation and fact verification. This work improves the reliability and ground truth of Large Language Models (LLMs), keeping outputs safe and factual.

I also organize and tag large datasets across text, images, and video on the Multimango platform to support multimodal AI systems. By maintaining high-quality multimodal annotation, I help models recognize and process different types of media together.

Previously, as a Content Moderator for AVID- AutoFlip, I reviewed over 150 video ads daily while enforcing Trust & Safety Protocols and content policy compliance. I consistently exceeded performance targets, achieved 120% production with >99% accuracy, and flagged policy violations to support risk mitigation (TAT>90%).