Zhang Xuanming

CrossCodeEval [AWS AI Labs] pointed out that LLM performs poorly in the code completion task with cross-file dependency calls. We designed dependency-chain-retrieval and mask-dual-library-retrieval to achieve SOTA performance.

A curated, structured knowledge base that systematizes exception types and handling logic across real-world programming environments. Supports training and evaluation of agents like Seeker for robust code synthesis.

Lightweight, intent-driven UI Agent framework for mobile environments with real-time multimodal interaction. Leverages Apple's native CoreML architecture for responsive user interfaces with EdgeSAM integration.

Novel network architecture based on Fourier Analysis, empowering the ability to efficiently model and reason about periodic phenomena by integrating Fourier Series into neural network structure.

Served as CTO of startup company, responsible for architecture design, function design and prototype construction. Received 500,000 RMB investment from Songshan Lake XbotPark Robot Base.

Optimization of SLAM point cloud VoxelGrid algorithm, training perception variation coefficient to construct homeomorphic mapping compatible with indoor/outdoor environments for autonomous navigation.