Bridging big data frameworks, AI platforms, and real-world business adoption, and seamlessly connecting with lakehouse data to deliver end-to-end data intelligence
Leveraging standardized components and development interfaces, it provides a one-stop development experience that helps enterprises and developers adopt algorithm engineering best practices.
Seamless Model Training with Big Data Systems
Reads structured and unstructured data directly from data lakes and data silos, seamlessly connecting data, feature preprocessing, and model training — eliminating tedious data import, export, and format conversion.
Sparse Feature Support
Applies to training scenarios such as large-scale sparse embedding layers, feature crosses, and variable-length feature pooling.
High-Performance Online Inference
Supports neural networks (including sparse embeddings), decision trees, and a wide range of traditional machine learning models, with accelerated heterogeneous hardware computing that lowers the engineering barrier to online deployment.
Unified Offline Feature Computation
A unified feature format and computation logic avoid duplicated development across systems and ensure consistency between offline and online features.
Online Algorithm Application Framework
Covers common online system capabilities such as automatic feature extraction from multiple data sources, feature computation, prediction service APIs, dynamic experiment configuration, and dynamic A/B test traffic splitting.
Embrace Open Source
We build core capabilities with multiple self-developed components while embracing the mature open-source ecosystem to lower the learning curve.
One-Stop Multimodal AI Platform
End-to-end coverage: a one-stop solution for diverse business problems
Business Application Framework
Bridges the last mile from algorithm models to business applications. Provides A/B testing, visual debugging, and hot parameter reload.
Online Inference Framework
A high-performance, low-latency model inference service for large-scale models. Supports heterogeneous hardware and can be deployed as a microservice.
Offline Training Framework
Bridges the gap between data and AI. Supports streaming online learning with higher efficiency and accuracy.
Solution: Low-Code Personalized Recommendations
BERT, Swing, Item CF, and other machine learning algorithms are tuned for different industries and massive datasets to deliver high accuracy and relevance. Non-AI professionals can deploy quickly with low-code tooling and improve key business metrics such as click-through and conversion rates.
E-commerce Recommendation
Overseas E-commerce Personalization
Import all your data quickly with just a few steps. The industry-grade recommendation system delivers personalized homepage experiences — You May Also Like, similar items, post-purchase recommendations, and more — helping users discover products faster and easier, creating a first-class shopping experience, driving purchase decisions, and growing your business.
