Skip to main content

Analytic Engines

Analytic Engines are the core processing components in ThirdLaw that can be configured to create specific Evaluators. Each engine provides different capabilities for analyzing and evaluating Exchange content.

What are Analytic Engines?

Analytic Engines are specialized processing components that analyze Exchange data and produce structured Findings. Each Engine uses different techniques and technologies to identify patterns, detect issues, or evaluate content against specific criteria.

When configured for a particular task, an Analytic Engine creates an Evaluator.

Available Analytic Engines

ThirdLaw includes four primary Analytic Engines, each specializing in different types of analysis:

Search Engine

The Search Engine looks for specific keywords and patterns in Exchange text. This engine is best suited for rapid detection of known patterns and structured text analysis.

Semantic Similarity Engine

The Semantic Similarity Engine focuses on identifying semantic relationships between content and reference collections. This engine goes beyond direct matching to understand broader semantic connections and topic relevance, and can be easily created from a Dataset of examples.

Validation Engine

The Validation Engine uses transformer-based zero-shot classification to validate content against predefined categories without requiring specific training examples. This engine excels at rapid, contextual detection of specific categories or patterns.

Foundational Engine

The Foundational Engine leverages LLMs to perform complex reasoning and analysis of Exchanges. This engine can identify subtle patterns and nuances that usually require human-like decision making, and can easily be customized for specific tasks via prompting.

Choosing the Right Engine

When developing an Evaluator, consider these factors for selecting the appropriate Analytic Engine:

Engine TypeBest ForExample Use Cases
SearchPattern matching, specific content detectionPrompt Injection, PII Detection, Code Detection
Semantic SimilarityThematic analysis, conceptual relationshipsCode Detection
ValidationContent validation, classificationToxic Language, Prompt Injection Validation
FoundationalComplex reasoning, contextual analysisToxic Language, Prompt Injection