Handshake AI
Video
Prompt + Response Writing (SFT)Managed the end-to-end data annotation pipeline for a Large Language Model (LLM) video-reasoning project. The scope involved high-density Prompt Response Writing and RLHF (Reinforcement Learning from Human Feedback) to improve model accuracy in visual temporal reasoning. Key tasks included: Authoring complex, multi-turn prompts to test model edge cases in video synthesis and descriptive accuracy. Performing comparative ranking of model outputs based on truthfulness, helpfulness, and safety constraints. Executing granular data labeling for thousands of video-text pairs, focusing on object persistence and motion dynamics. Adhered to strict 98%+ inter-annotator agreement (IAA) standards and utilized iterative feedback loops to refine the model's linguistic nuance.