At Alignerr, I evaluate multi-turn LLM responses and annotate speech-turn boundaries in voice activity detection clips, supporting conversational alignment and ASR accuracy. At Turing, I annotated robotics and egocentric video data, maintaining 97% accuracy across object interaction, spatial reasoning, and multi-step task sequences.
At Xlairs, I engineer audio-visual reasoning prompts for long-video datasets and compare model predictions with ground-truth answers to identify hallucinations and multimodal inconsistencies. At Soul AI, I create audio-to-text prompts and score model responses using a five-dimension rubric.
