Multi-Agent System
We designed specialized AI agents to handle different aspects of procurement:
- Supplier Evaluation: Verifies transaction history and ratings
- Logistics Optimization: Calculates real-time shipping costs
- Compliance Checker: Screens against export control lists
Key Update: In February 2025, we enhanced this system by integrating DeepSeek's inference models for improved decision support in complex scenarios.
AI Model Stack
| Layer | Components | Functionality |
|---|---|---|
| Foundation | Qwen2.5 + Custom Algorithms | Cross-language search optimization |
| Enhancement | DeepSeek Models | Advanced decision support |
This hybrid approach allows us to:
- Process queries in 5 languages (EN/DE/FR/ES/PT)
- Understand industrial terminology (e.g., "ASME B16.5 flanges")
Data Infrastructure
Our real-time processing handles:
- 250,000+ supplier profiles
- 30M+ cross-border trade relationships
- Continuous updates via ERP system CDC pipelines
Performance Highlights
- Query Response: <30 seconds for complex procurement plans
- Accuracy: 20-30% higher conversion rates than traditional search (per our 2024 tests)
- Scalability: Serves 1M+ enterprise users as of March 2025
Lessons Learned
- Industrial NLP: Requires heavy domain adaptation
- Multi-Lingual Support: Business terminology differs significantly from general language
- Real-Time Data: Supplier information has shorter validity than consumer goods data
For fellow engineers: How would you approach the challenge of standardizing industrial part numbers across languages?
Top comments (0)