Gaotu wanted to scale capabilities such as automatic homework grading and intelligent Q&A, but needed a stable large-model architecture that could handle education-specific accuracy, privacy, integration, and real-time response challenges.
Reported impact
Gaotu reduced teacher workload, improved student support responsiveness, and created a more sustainable technical base for adding new AI education features across its course portfolio.
Gaotu and Baidu AI Cloud shaped a large-model operating approach for online education rather than a single isolated feature. The project uses Qianfan's model-development and application platforms as the basis for automatic grading, intelligent Q&A, digital-human teaching support, and teaching-task design. Around that, Gaotu and Baidu addressed the practical constraints of education AI: model quality is improved through pretraining and fine-tuning strategies plus performance monitoring; compute cost is controlled through model compression, pruning, distributed training, and optimization algorithms; and data quality is protected through standardized collection, cleaning, privacy controls, and semi-automated labeling workflows. On the runtime side, configurable interfaces, caching, asynchronous processing, and fault-recovery design help the services stay responsive as requests grow. The result is an education-oriented large-model foundation that lets Gaotu connect tutoring, assessment, and teaching operations to one evolving AI stack.
Student work, tutoring questions, and teaching resources are processed through Qianfan-based models that perform grading, question answering, and content support. The same stack uses standardized data pipelines and optimized runtime controls so education services can return faster responses while preserving privacy and enabling ongoing model improvement.
Baidu and Gaotu addressed technical challenges through pretraining, fine-tuning, monitoring, compression, pruning, and distributed training (original: 引入预训练和微调策略，建立性能监控体系…采用模型压缩和剪枝技术，并基于百度智能云的平台优势，探索分布式训练).
Data collection, cleaning, privacy protection, automated labeling, and review mechanisms are all part of the solution (original: 建立严格的数据采集和清洗流程，加强数据隐私保护…引入自动化标注工具，建立专业标注团队和审核机制).
Gaotu is using large models for automatic homework grading and intelligent Q&A (original: 高途利用大模型实现作业自动判题、智能问答等业务能力).
Runtime performance is improved with configurable parameters, caching, asynchronous processing, fault recovery, and load balancing (original: 开发可配置模型参数和接口…采用缓存技术和异步处理机制，建立故障恢复和负载均衡机制).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Baidu AI Cloud Sample Oct 13, 2025 Kaoshibao (Shanghai Juxian Network Technology Co., Ltd.) Kaoshibao needed to use new AI interaction and content technologies to improve personalized learning, lower content-production costs, and make very large question banks easier to ingest, search, and explain. "Zero-basis" users improved certificate pass rates 35%. Intelligent Search and Generation Component · Multimodal Recognition · Qianfan AppBuilder Baidu AI Cloud Sample Jun 23, 2025 Tongji University Tongji University needed a domestic AI stack with stronger computing and model-tooling support for embodied intelligence, multimodal models, autonomous perception and multi-agent research. BaiGe AI Heterogeneous Compute Platform · Kunlun P800 · Qianfan AppBuilder Baidu AI Cloud Sample May 15, 2025 Lanxum Education (立思辰) Lanxum wanted to make Chinese-language learning more engaging and personalized, but it needed a structured way to connect teaching content, teacher knowledge, dialogue, and student guidance instead of relying on static… Chinese Teaching Research Knowledge Graph · Intelligent Voice Dialogue System · Virtual Teacher 