Project
has rich industry project engineering experience
Digital Twin Platform
Industrial-grade digital twin system based on MB-TwinEngine architecture, integrating NVIDIA Omniverse and Unity3D stacks. Features: 1M+ data point real-time mapping (120k I/O signals/sec); multi-physics coupled simulation accuracy >99.3%; 300+ standard production line templates; WebGL 3D rendering (5x faster loading).
Intelligent Data Annotation Platform
Features: Supports 200+ annotation types (2D/3D/time-series); AI pre-annotation accuracy >92%; collaborative efficiency +300%; AutoQC evaluation (F1-score 0.95+); COCO/VOC/KITTI export. Daily capacity reaches 500k units with consistency error <0.3%.
AI Agent Editing Platform
Multimodal agent development environment based on MB-AgentFramework, supporting zero-code visual editing and 100B-parameter model deployment.
AI Coding Assistant Platform
Integrating GPT-4 and CodeLlama-70B stacks. Core features: Code generation across 20+ languages (Python/Java/Go); 300+ industry coding standards; distributed A100 compilation; AST parsing accuracy >98.5%. Daily generation capacity reaches 800k lines with AutoDebug fixing 90%+ syntax errors.
AI Agent Parameter Tuning Workshop
Features: Fine-tuning for 100B+ models (H100/A100 multi-GPU); automatic mixed-precision training (FP8 saves 50% VRAM); gradient accumulation & checkpoints (recovery <30s); TensorBoard monitoring. Yields 15.2% higher accuracy with 8x faster tuning than traditional methods.
Ragflow RAG Platform
3rd-gen RAG system based on MB-RAGFramework, combining BM25 and deep semantic matching. Real-time multi-source data indexing (200+ formats); hybrid search (keyword/vector/semantic); 95%+ recall rate in key scenarios with FP16 compression saving 70% VRAM.