NimbleEdge Platform
Effortlessly deploy and maintain real-time machine learning models on mobile edge for session-aware, privacy-preserving personalization
Benefits
NimbleEdge Cloud Orchestration
NimbleEdge
Simulator
Run large scale ML inference and training simulations to estimate uplift benefits from real-time personalization
NimbleEdge
Cloud Service
Managed service orchestrating the machine learning pipelines for edge deployment
NimbleEdge
SDK
Ultra light-weight SDK integrated into client mobile app to enable on-device ML capabilities
NimbleEdge Cloud Service
Serializes and hosts ML models to be downloaded by mobile edge device
Orchestration engine to determine execution and data flow management
ML teams have complete central control of the on-device pipelines, including on-the-fly updates of data processing pipeline and model updates
Pre-built integrations with AWS, Azure, and Databricks for leveraging existing models and cloud feature stores
Learning Platform
Aggregates on-device individualized models to update (train) models
Ensures model training while Personally Identifiable Information (PII) remains on users’ mobile devices
NimbleEdge SDK
Warehouse
On-device managed database for storing persistent raw data such as real-time product-user interaction events
On-device managed data processing engine for computing real-time user interaction based aggregate features
Ultra fast feature computation for each user by virtue of decentralized processing
Feature Store
On-device in memory processed data replica store for dynamically computed feature vectors from user events and cloud-based data
Precompute features within 80 microseconds, ~1000x faster than central cloud feature stores
Inference Engine
OS, CPU architecture and model specific compile and run time optimizations for minimum latency and resource consumption
End-to-end latency of <50 milliseconds from user interaction to prediction
Unique model for each user for hyper-personalized results
5-10% uplift in model performance
Privacy-aware ML as PII is not transmitted to cloud servers for model training