Global And Entity-Specific Deployment Engines for Low-Latency Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional contact center systems face significant latency issues due to the high computational demands of matching natural language queries to hundreds or thousands of intents, as they rely on centralized approaches that do not efficiently deploy resources for specific intents.
Innovation Solution
Implementing a distributed computing framework where queries are processed by multiple deployment engines, each associated with a specific intent, and a controller engine dynamically assigns computing resources to these engines based on demand, allowing for real-time intent matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a centralized approach is used for intent matching, then system simplicity is maintained, but latency increases due to high computational demands
Solution Approach 1:
The patent divides the centralized intent matching system into multiple distributed deployment engines, each handling specific intents. This segmentation distributes computational load across multiple nodes, reducing latency for query processing while maintaining system functionality through modular architecture.
Solution Approach 2:
The system transitions from a single-dimensional centralized processing model to a multi-dimensional distributed architecture. Deployment engines are distributed across multiple dimensions (spatial distribution, functional specialization), enabling parallel processing that reduces latency without proportionally increasing perceived system complexity.
2Productivity
If computing resources are statically allocated, then resource management is simple, but efficiency decreases due to uneven demand across intents
Solution Approach 1:
The patent implements dynamic resource allocation where the controller engine continuously monitors computational demand across deployment engines and adjusts resource distribution in real-time. This dynamic approach optimizes processing efficiency by allocating more resources to high-demand intents while reducing allocation to low-demand intents, preventing resource waste.
Solution Approach 2:
The system incorporates feedback mechanisms where deployment engines report their computational load and performance metrics to the controller engine. This feedback loop enables the controller to make informed decisions about resource allocation, adjusting distributions based on actual demand patterns observed in the system.
3Reliability
If all queries are processed by all deployment engines, then comprehensive matching is achieved, but computational overhead increases significantly
Solution Approach 1:
The patent assigns specific intents to specific deployment engines based on their specialized capabilities and knowledge. Each deployment engine develops local expertise in its assigned intents, enabling accurate matching for those specific intents while avoiding the computational overhead of universal processing across all engines for every query.
Solution Approach 2:
The system extracts and isolates specific intent processing functions into separate deployment engines. This extraction allows queries to be routed to only the relevant subset of engines that handle specific intents, eliminating unnecessary computational energy spent on engines that cannot match the query intent.
Data Source
AI summary
An entity associated with a query received from a user device at a contact center server is identified. The query is transmitted to a first set of deployment engines associated with global intents common to multiple entities and to a second set of deployment engines associated with entity-specific intents for the identified entity. Scores are received from the first set of deployment engines and from the second set of deployment engines. Each score represents a likelihood that the query matches to an intent associated with the respective deployment engine. A subset of intents for which the likelihood exceeds a threshold is aggregated. A prompt is transmitted to the user device to select from the aggregated subset of intents.