Topic

The Road to AI Native R&D: Kwaipilot's Practical Implementation in Racer

Kwaipilot applies Racer's self-developed code generation model, which is the first time in China that the MoE architecture is successfully applied to code continuation modelling and has achieved significant gains, and provides three major products: IDE plug-in, Q&A engine, and intelligent body application development platform. Kwaipilot has realised the data flywheel of ‘coding is annotation’, and with the effect of data-enhanced model, it has surpassed the GPT-4 by 10% in the semantic understanding and generation scenario of Kwaipilot's dialect code. This session will introduce how Kwaipilot AI R&D tool product explores the whole life cycle scenario of R&D on Racer, and explains the core principles of AI native R&D tool product design driven by developer experience. Outline ● Evolutionary history and development trend of AI R&D tool products and industry. ● The exploration and practice of the whole bucket of Kwaipilot AI R&D tools in improving efficiency in the whole R&D lifecycle of Racer: ○ Introduction of Kwaipilot product matrix and the current situation of implementation The core principles of developer experience-driven AI native R&D tool product design. ○ Specific practices and cases of how Kwaipilot applies the above principles, and the real value of innovation in actual user scenarios. Kwaipilot's continuous exploration and outlook in the field of AI R&D tool products: future software development trends and exploration of next-generation productisation.