xDAN AI

Intern Team Leader · March–June 2024
Supervisor: Yipeng Cheng

• Synthetic Data Generation: Developed LLM-based data augmentation pipelines across 10+ datasets, including poetry and other domain-specific tasks, to improve the diversity and coverage of instruction-tuning data.
• Data Quality Evaluation: Developed an LLM-based data scoring framework inspired by DEITA and UltraFeedback to evaluate sample complexity and quality, supporting the construction of high-quality instruction-tuning datasets.