Amphigraph

A 156-feature classification engine for industrial maintenance that links every AI decision to source data. Combines Datalog reasoning, LLM consensus, and knowledge graph validation.

The Problem

Industrial maintenance organizations struggle with unstructured work order data. Traditional NLP approaches fail because they can't explain their decisions—a critical requirement in regulated industries.

The Solution

Amphigraph uses a hybrid architecture:

• Datalog Reasoning (Soufflé) for deterministic rule application
• LLM Consensus for ambiguous cases requiring judgment
• Knowledge Graph Validation (Neo4j) for relationship verification

Every classification links back to source evidence, enabling auditable AI decisions.

Results

• 500K+ work orders processed
• 156 features shipped to production
• ISO 14224 compliant taxonomy