LASEV is a hierarchical LLM-based multi-agent system for generating high-quality educational videos from problems. Rather than end-to-end pixel synthesis, our system constructs structured executable video scripts that are deterministically compiled into synchronized visuals and narration.

Multi-agent system architecture with Orchestrating Agent coordinating working agents

• 🤖 Multi-agent architecture with specialized working agents for reasoning, visualization, and narration
• 🎓 Designed specifically for educational content generation
• 📊 Throughput exceeding 1M videos per day
• 💰 95%+ cost reduction compared to industry standards
• ✅ High acceptance rate in production deployment

If you find this work useful, please cite:

@misc{yan2026lasev,
  title={Beyond End-to-End Video Models: An LLM-Based Multi-Agent System for Educational Video Generation},
  author={Yan, Lingyong and Wu, Jiulong and Xie, Dong and Shi, Weixian and Xia, Deguo and Huang, Jizhou},
  year={2026},
  eprint={2602.11790},
  archivePrefix={arXiv}
}