Ze Chen
Baidu Wenshin Express Code Comate Engineer
Baidu senior front-end engineer, internal lecturer, DevOps iCafe product leader, with more than 5 years of experience in software engineering efficiency, was responsible for the implementation of Baidu's internal DevOps project cloud native transformation, internal NPM mirror leader. 2023, participated in the development of Baidu's IDE plug-in for the intelligent development tool, Comate, and was responsible for the RAG retrieval-based dialogue enhancement of large models. Big Model Dialogue Enhancement, reducing code vector index building time consuming from minutes to seconds. He is good at performance optimisation of web applications and has rich experience in LLM application development.
Topic
AI Reinventing Development: An Engineer's Perspective on the Growth Journey of MindFast Code Below
From an engineer's point of view, we share the evolution of Wenshin Express from internal project to growing into a leading coding plugin in China. In this process, the continuous iteration of the big language model drives the updating of the technical reserves: from the basic streaming output experience, Function Call, Prompt engineering, to the early model training, evaluation and fine-tuning, to building the plug-in ecosystem, knowledge management, RAG retrieval, and up to the latest practice of intelligent body application, every progress of Wencent Express Code can not be separated from the accumulation of these technologies. Accumulation. Meanwhile, as users of Wenshin Express Code, we are constantly thinking and deciding which features are worth being developed, exploring the future blueprint of programming plug-ins empowered by AI, so that AI can truly become an efficient collaboration partner for engineers. Outline: 1.The origin of tools: the original intention of Comate (coding mate) 2.The iPhone Moment of Big Models: Reconstructing Products with AI Native Thinking -Initial Practice: Training and Deploying Open Source Models to Achieve Natural Language Transformation into Custom Query Languages Scenario landing: using Function Call to solve the long-tail requirements of web applications based on Wenxin Yiyin. -Exploration: using AI models to achieve automatic requirement realisation, and AutoWork forms appear. 3.Productisation 1.0: the blueprint of coding plug-ins is seen, and the code renewal ability evolves. 4. Productisation 2.0: -How to establish an open ecosystem to help enterprise development efficiency 15% -RAG practice: establishing a user-oriented knowledge management platform -Explore dialogue enhancement for programming scenarios in various languages 5. Productisation 3.0: -Embedding Practice: Application of Vector Retrieval in Dialogue Scenarios -Terminal Problem Repair: Contextual Enhancement-based Problem Repair Practice 6. Outlook: Intelligent Body Practice, AI Delegated Programming Tasks
