Xiaomi Credit Intelligence & Risk Platform
Built a credit-scoring platform from scratch for thin-file users, converting MIUI, IoT-device, and e-commerce behavior into 1,500 features, an XGBoost model, and API-based risk services.
Problem & Constraints
Thin-file users lacked traditional credit history, leaving financial institutions without enough signals for reliable risk decisions. The solution needed interpretable features, production-scale scoring, partner-facing APIs, and flexible commercial service tiers.
Engineering Role
Built the platform, engineered and ranked the feature system, trained the XGBoost model, designed the API services, and supported product tiers and joint-modeling workflows for financial institutions.
System Architecture
Technical Approach
- Engineered features from MIUI usage, IoT devices, and e-commerce behavior
- Ranked signals using Information Value and trained an interpretable XGBoost model
- Served risk scores through standardized APIs to external institutions
- Designed three B2B credit-tag tiers plus a joint-modeling service
- Extended the underlying data assets into user profiling and marketing activation capabilities
Production Impact
Reached AUC 0.75+, served 500M+ scoring calls and 2M+ users, supported 40+ financial institutions and 20B+ RMB in lending, drove 500K+ new users and 50M+ RMB in revenue for card and payment teams, and improved partner risk-approval rates by 10–20%.
What This Proves
Demonstrates end-to-end production ML engineering: feature systems, interpretable modeling, APIs, scale, partner integration, and measurable financial impact.
Ready to discuss role fit?
If this system maps to your AI engineering, enterprise delivery, data, or risk needs, reach out directly.