Document Processing
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.
Reported impact
Kaoshibao cut labor and content-production costs, improved search and explanation quality, boosted course completion and pass rates, and increased paid conversion with new AI learning features.
Kaoshibao used Baidu AI Cloud's Qianfan platform to turn a large exam-content operation into an AI-native learning and content-production pipeline. Qianfan ModelBuilder supplied the large-model capabilities behind question generation, explanation, key-point extraction, and multimodal recognition, while Qianfan AppBuilder assembled the user workflows for photo question search, AI question import, and the learning assistant. Uploaded Word, PDF, and image files pass through multimodal recognition to extract text and numbers, then are normalized into structured question-bank assets with deduplication, subject tagging, question-type classification, and difficulty grouping. An Intelligent Search and Generation Component grounds explanations on Baidu question-bank, Wenku, and Baike sources, while Video Understanding Technology turns long lessons into concise notes and knowledge maps that feed tutoring, search, and paid learning services.
Users upload Word, PDF, and image files or submit photo and video learning queries into Qianfan AppBuilder workflows. Multimodal Recognition extracts and structures question content, Qianfan ModelBuilder generates answers, explanations, and key points, the Intelligent Search and Generation Component grounds those outputs on Baidu content sources, and Video Understanding Technology produces condensed notes and knowledge maps that flow back into Kaoshibao's question-bank and tutoring experiences.
A 2-hour video lesson can be condensed into 15 minutes of illustrated notes plus a knowledge map, raising completion to 89% and improving certificate pass rates for 'zero-basis' users by 35% (original: 将2小时视频课浓缩为15分钟图文笔记+知识脑图，完课率提升至89%；使“零基础”用户专业证书通过率提升35%).
Hundreds of millions of question explanations were reduced from months to within two weeks (original: 数亿题解析耗时从数月缩至2周内).
Kaoshibao used Baidu AI Cloud's Qianfan platform and ERNIE large-model technology to add photo question search, AI question import, and AI explanation capabilities (original: 通过百度智能云千帆平台引入文心大模型技术，实现拍照搜题、智能导题、AI讲题等创新功能).
Paid conversion increased 25% (original: 付费率提升25%).
The intelligent search and generation component relies on sources such as Baidu's question bank, Baidu Wenku, and Baidu Baike to enrich explanation content at very low cost (original: 智能搜索生成组件依托百度题库、百度文科、百度百科等多种权威可靠信源，支持以极低成本更新和丰富题目解析内容).
The platform's question-bank volume has exceeded 3 billion questions (original: 平台题库总量已突破30亿道).
Through deep optimization and multimodal verification on the Qianfan platform, text and numerical recognition accuracy approached 100%, labor costs fell 99.8%, and content-production efficiency improved 1000x (original: 通过百度智能云千帆平台的深度优化与多模态校验，文本/数字识别准确率接近100%，整体人力成本下降99.8%，内容生产效率提升1000倍).
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
Microsoft Sample Oct 15, 2026 Bar-Ilan University Students who miss lectures struggle to reconnect new lessons with prior material when course information is spread across notes, recordings, slides, and readings, and lecturers need a reliable way to answer routine… 71 courses Azure AI Services · Microsoft Azure · Microsoft Copilot Studio 
IBM Sample Jan 2, 2026 National Agency for Education, Lithuania Lithuania’s National Agency for Education needed to extract, standardize, and compare information from thousands of mixed-format school management reports because manual extraction took hours per report and made… 14 key criteria tracked by AI solution IBM Cognos Analytics · IBM watsonx.ai IBM Sample Sep 15, 2025 The University of Auckland University of Auckland researchers were spending too much time navigating grant, ethics, and export-control requirements because funding applications and compliance reviews involved multiple forms, committees, and rules… 85% accuracy in detecting known error types in sample applications IBM Cloud · IBM watsonx Assistant · IBM watsonx.ai 
