Model Selection & Configuration

Protege AI's AI system allows you to change the default AI model for your content type and review requirements. Our proprietary ShieldLlama model, combined with specialized options, provides optimal accuracy for compliance review.

Configuration in Policy Steps

Model selection is simplified - the system handles complexity automatically:

Policy Step Setup

1. Navigate to Policy Engine → Your Policy Group → Edit Step

2. Model Configuration section shows:

   • Primary Model: Automatically selected (typically ShieldLlama)

   • Specialized Options: Available if needed for your content type

   • Vision Processing: Auto-enabled when there is a Prompt in Vision Prompt

Available Models

Recommended: ShieldLlama (Proprietary)

Our flagship model, optimized specifically for compliance and content review:

• Best for: All standard compliance, marketing review, and policy enforcement

• Advantages: Purpose-built for your use cases with superior accuracy, trainable with Teach your AI

• Performance: Fastest processing with highest precision for compliance content

• Default choice: Automatically selected for most review scenarios

Specialized Models

Qwen 235 A22B Instruct

• Best for: Large document processing (emails, lengthy reports, multi-page documents)

• Advantages: Exceptional handling of long-form content and complex document structures

• Use cases: Employee email review, comprehensive document analysis, multi-section policy documents

Llama 3.3 70B

• Best for: Reasonable processing requiring specific logic

• Advantages: Superior analytical capabilities for nuanced policy interpretation

• Use cases: Edge case analysis, complex regulatory interpretation, multi-step compliance decisions

Llama Maverick & Scout

• Maverick: Advanced general-purpose model with broad capabilities

• Scout: Optimized for rapid initial screening and triage

• Use cases: Custom workflows requiring specific processing characteristics

Vision Models

Llama Maverick Vision (Default)

• Automatically selected for all image and video content

• Capabilities: Image analysis, OCR, visual content compliance review

• Handles: Screenshots, marketing images, social media content, document scans

Specialized Vision Models

• Available on request for specific use cases

• Contact support to enable advanced vision capabilities

• Custom configurations for industry-specific visual content

Model Recommendations by Use Case

Standard Marketing Review

✅ ShieldLlama (Default)

• Social media posts

• Marketing copy

• Website content

• Advertising materials

Employee Communication Review

✅ Qwen 235 A22B (Recommended for large documents)

• Email chains and threads

• Internal communications

• Lengthy reports and memos

• Multi-page documents

Complex Policy Decisions

✅ Llama 3.3 70B

• Edge case interpretation

• Multi-step compliance analysis

• Regulatory gray areas

• Appeals and escalations

Visual Content Review

✅ Llama Maverick Vision (Default)

• Marketing images

• Social media visuals

• Document screenshots

• Video content frames

High-Volume Screening

✅ ShieldLlama + Scout

• Bulk content processing

• Initial triage and filtering

• Real-time content monitoring

• Automated compliance checks