Project Background & Challenges
Shuxingyun Tech recently deployed an enterprise AI Agent platform for a major Chinese financial group, marking a significant milestone in applying domestic LLMs to mission-critical finance workloads.
Before deployment, the client faced several critical challenges: slow customer service response (averaging 8,000+ daily tickets), lagging risk control rules (200+ new rules added monthly were hard to maintain), and time-consuming research report generation (each report required 2-3 days). The client sought to upgrade workflows using LLM capabilities and AI Agent collaboration.
Technical Solution & Core Architecture
The platform integrates Tongyi Qianwen, DeepSeek, and Wenxin Yiyan, building a "one-brain multi-modal" intelligent agent system. The architecture has three layers:
Model Layer: Unified routing automatically selects the best model per scenario (Tongyi Qianwen for dialogue, DeepSeek for code, Wenxin Yiyan for risk reasoning).
Agent Layer: 8 vertical agents built on RAG knowledge base and Function Call, handling customer service, risk control, reporting, research, and other tasks. Supports multi-agent collaborative workflows.
Application Layer: Connects to 12 internal business systems, enabling automatic ticket routing, real-time risk interception, and one-click report generation.
Core Capabilities & Quantitative Results
After deployment, business processing efficiency improved by 60%, with manual ticket volume dropping 45%. Risk control response time shrank from hours to seconds, with risk identification accuracy reaching 95%. Research report generation compressed from 2-3 days to 2-4 hours, with report quality scores rising 30%.
Client feedback: the AI Agent platform has become an indispensable intelligent assistant for business departments. Next steps will expand into investment advisory, regulatory reporting, and intelligent marketing scenarios.
Delivery & Future Outlook
The project was delivered over 6 months in three phases. After 30 days of stable operation, the platform handles 50,000+ business requests daily.
Next plans: 1) integrate Qwen3 and DeepSeek-V3 next-gen models, 2) expand agents to 15, 3) explore multimodal capabilities for chart understanding and PPT auto-generation. Shuxingyun will continue partnering with the client to scale AI Agent adoption in finance.