Platform-Agnostic Semantic Search with Split Training and Inference

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing semantic search systems face limitations due to the use of a single platform for both training and inferencing, which can lead to suboptimal performance in terms of training time and accuracy during inferencing.

Innovation Solution

Implementing a two-platform approach where a first platform (e.g., Python) is used for training and a second platform (e.g., Java) is used for inferencing, utilizing machine learning models to generate training artifacts and perform semantic searches, respectively.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a single platform is used for both training and inferencing, then the system structure is simple, but the training time and inferencing time are both suboptimal

Engineering Contradiction:

Improvetraining timeVSAvoidsystem structure

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system is segmented into two distinct platforms: a first platform (e.g., Python) dedicated to training machine learning models, and a second platform (e.g., Java) dedicated to inferencing. This segmentation allows each platform to be optimized for its specific function, reducing training time by 68% and inferencing time by 50%, while maintaining manageable system complexity through clear functional separation.

Inventive Principle:

Principle #1Segmentation

2Manufacturing precision

If a single platform is used for both training and inferencing, then the system is easy to operate, but the accuracy during inferencing is suboptimal

Engineering Contradiction:

Improveinferencing accuracyVSAvoidsystem operation

Core Design Contradiction:

Manufacturing precisionVSEase of operation

Solution Approach 1:

By separating training and inferencing operations onto different platforms, each platform can be independently optimized. The first platform focuses on model training with appropriate libraries and frameworks, while the second platform focuses on high-accuracy inferencing with optimized runtime environments. This improves inferencing accuracy by 15% while maintaining ease of operation through standardized interfaces and automated model deployment.

Inventive Principle:

Principle #1Segmentation

3Productivity

If a two-platform approach is used, then training time and inferencing time are reduced, but the system complexity increases

Engineering Contradiction:

Improveinferencing timeVSAvoidplatform integration

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

An intermediary layer is introduced to bridge the first platform (training) and second platform (inferencing). This intermediary handles model serialization, artifact transfer, and interface standardization, allowing the two platforms to communicate efficiently. This reduces inferencing time by 50% while keeping platform integration complexity manageable through well-defined communication protocols and standardized data formats.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250307291A1Platform agnostic scalable and high-performance semantic search framework

Publication Date: 2025.10.02 SERVICENOW INC

AI summary

Various implementations disclosed herein include performing a semantic search for records related to an input query according to a similarity score identified using a second machine learning model of a second platform based on training artifacts generated using a first machine learning model of a first platform.