免费领取大会全套PPT    

点击领取

我要参会

Yusen Zhao

NetEase CodeWave Intelligent Development Platform Architect with 10 years of software development experience. Responsible for the overall architecture design of the low-code visual programming language (NASL) and full-stack application framework, as well as the research and implementation of the integration of low-code programming and AI technology.

Topic

Low-Code Platform Deeply Integrated with AI

In the traditional development domain, the AI programming tool industry is booming. Code generation, code completion, and other auxiliary tools have made programming more efficient and full-stack. Tools like Cursor, Replit, and v0.dev are redefining IDEs, while traditional Design-to-Code (D2C) technologies are achieving new breakthroughs when combined with large models. These technologies pose certain challenges to low-code products, which are also efficiency-enhancing tools, but they also bring new opportunities and transformations. As a technical team that has been one of the first in China to extensively utilize the capabilities of large models and implement them in products, this presentation will start from the current status of the development field and the nature of low-code products themselves. It aims to analyze the opportunities and difficulties of integrating low-code with AI, introduce product design and technical architectures that combine various programming domains such as pages, logic, and styles with code generation, code completion, D2C, and other AI technologies. Furthermore, it will delve into practical experiences with data-driven language model training and mechanisms for product iteration and updates. Outline: Development and Bottlenecks of Low-Code and AI Integration Language-Based Low-Code AI Service Architecture Base Design of Visualization Programming Languages for Streamlining Tech Stacks Multi-Stage Generation Pipeline for Robustness Multi-Agent Collaborative Architecture with Integrated Retrieval-Augmented Generation (RAG) Capabilities Practical Case Study: Natural Language Routing, Generation of Visual Logic and UI Data-Driven Training of Low-Code Language Models Product Iteration Update Mechanism Driven by Benchmarks Construction of an AI Engineering Platform for Efficient Model Training: Metric Statistics, User Data Feedback, Online Switching and Disaster Recovery of Models Practical Case Study: Context-Specific Enhancement of Code Completion, Multi-Modal Progressive Enhancement of D2C (Design Drafts/Screenshots to Low-Code Pages) Summary and Outlook