Hardik Singh Sengar

Annotated and labeled images and video frames for computer vision model training using bounding boxes, polygons, and semantic segmentation masks. Performed quality checks and review passes to ensure accuracy and guideline consistency. Identified edge cases such as occlusions and ambiguous objects, and flagged them with detailed notes to refine labeling taxonomies and improve training data quality. Collaborated with ML engineers to improve annotation guidelines and reduce labeling errors.

Conducted advanced LLM evaluation and response quality analysis across multiple AI workflows. Supported RLHF pipelines including SxS evaluation, grounding analysis, and hallucination detection. Designed and refined prompts for multi-turn interactions to improve accuracy and reliability. Provided structured feedback and annotations to enhance model performance, safety, and compliance.

Train and fine-tune large language models by providing high-quality data annotations, feedback, and evaluations to enhance model accuracy and performance. Analyze AI outputs to identify edge cases, biases, hallucinations, and personalization errors. Conduct Side-by-Side (SxS) evaluations and grounding analysis to improve response integration and helpfulness. Generate structured rationales and detailed quality reports to strengthen model reliability and safety compliance.