Foundation Model Advisor for Generative AI Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The proliferation of foundation models in various hubs and repositories makes it challenging for users to select the most appropriate models for client-specific generative AI solutions, often requiring significant user effort and resulting in incorrect outcomes, increased costs, and time spent.

Innovation Solution

A computer-implemented method and system that uses a foundation model advisor to generate a list of recommended foundation models, ranked based on attribute matching scores and accompanied by justifications, which are then adjusted based on user feedback to ensure accuracy and appropriateness for specific generative AI needs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If users manually select foundation models from multiple hubs and repositories, then they can choose models for client-specific generative AI solutions, but it requires significant user effort and results in increased time and cost

Engineering Contradiction:

Improveease of foundation model selectionVSAvoidtime spent on model selection

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The foundation model advisor system performs self-service by automatically selecting and ranking foundation models based on client-specific requirements without requiring manual user intervention. The system autonomously evaluates models against criteria such as performance, cost, and suitability, thereby eliminating the time-consuming manual selection process while maintaining high accuracy in model choice.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The foundation model advisor acts as an intermediary between users and the vast array of available foundation models. It translates client-specific requirements into model selection criteria, evaluates numerous models objectively, and presents ranked recommendations with justifications. This intermediary function significantly reduces user effort and time while ensuring appropriate model selection.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If users manually select foundation models without automated assistance, then they can make selections, but it increases the risk of incorrect outcomes and costs

Engineering Contradiction:

Improveaccuracy of foundation model selectionVSAvoiduser effort required

Core Design Contradiction:

ReliabilityVSEase of operation

Solution Approach 1:

The foundation model advisor incorporates feedback mechanisms that continuously learn from user selections and outcomes. By analyzing feedback data, the system refines its recommendation algorithms to improve accuracy over time. This feedback loop ensures that automated selections become increasingly reliable while reducing the need for user intervention.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system replaces manual mechanical selection processes with automated AI-based evaluation mechanisms. The foundation model advisor uses machine learning models and automated criteria to assess foundation models, substituting human judgment with objective, data-driven algorithms that reduce selection errors and costs.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If automated foundation model recommendation is implemented, then selection accuracy improves, but the system complexity increases

Engineering Contradiction:

Improveaccuracy of foundation model selectionVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The foundation model advisor system is segmented into distinct functional modules: requirement analysis module, model evaluation module, ranking module, and justification generation module. Each module handles a specific aspect of the selection process independently, making the overall complex system more manageable and easier to implement while maintaining high accuracy through specialized processing at each stage.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12314321B1Automated selection of foundation models for client-specific generative artificial intelligence needs

Publication Date: 2025.05.27 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Automatically recommending appropriate foundation models is provided. A list of recommended foundation models, along with a corresponding ranking for each respective foundation model in the list is generated based on a corresponding attribute matching score and a justification for the corresponding ranking of each respective foundation model. The list of the recommended foundation models, along with the corresponding ranking for each respective foundation model in the list and the justification for the corresponding ranking of each respective foundation model is sent to a user. Feedback is received from the user regarding accuracy of a user-selected foundation model from the list of the recommended foundation models. The user-selected foundation model is adjusted based on the feedback received from the user regarding the accuracy of the user-selected foundation model from the list of the recommended foundation models.