aiXcoder Deploys Custom Large Model for Enterprises, Code Generation Accuracy Rises to 45%

Phoenix Net Technology News August 13th. Recently, a leading telecommunications company in China has deployed a custom large model through aiXcoder's tailored solutions, significantly improving the efficiency of its research and development scenarios. The model achieved a 25% increase in code generation accuracy (reaching 45%) and an 18% increase in enterprise knowledge question-answering accuracy (reaching 69%), breaking through the bottlenecks caused by general models lacking private domain knowledge.

Prior to that, the company introduced a general large model to assist in research and development, but due to the lack of proprietary protocol stacks, equipment interaction logic, etc. private domain knowledge, it encountered issues such as question-answering failure, low code usability, etc. aiXcoder achieved deep optimization through four major technical paths: adopting a "large model + small model" collaborative strategy to strengthen professional scenario processing capabilities; building an end-to-end private domain data governance system to bridge the knowledge gaps in research and development workflows; integrating workflow and Agent technologies to fill gaps in end-to-end capabilities; and using AI-generated high-quality training data to solve data volume bottlenecks.

The framework formed by this solution possesses high reusability, supporting enterprises to quickly iterate data and models based on business needs, reducing subsequent training costs. The practice has already taken shape in industries such as finance and aerospace.