Multi-Agent RCA
When an anomaly is detected, a LangGraph-orchestrated pipeline of 4 specialised AI agents automatically investigates the fault, traces its root cause, and produces an actionable maintenance plan.
The 4-agent pipeline
Agents execute in sequence. Each feeds its output into the next, building context progressively.
Diagnostic Agent
Symptom analysisReceives the raw anomaly signal and ensemble score. Classifies the symptom cluster and queries the knowledge graph for matching failure patterns.
Reasoning Agent
Root cause inferenceUses Groq Llama 3.3 70B to reason over the symptom cluster, equipment history, and SWRL rules. Returns ranked probable root causes with confidence scores.
Planning Agent
Maintenance recommendationsTranslates confirmed root causes into prioritised maintenance tasks — from immediate corrective actions to scheduled preventive measures.
Learning Agent
Continuous improvementRecords technician feedback and outcome data to refine future diagnoses. Accuracy improves with every confirmed or corrected RCA.
Technology
LangGraph
Directed agent orchestration with state management and conditional branching.
Groq Llama 3.3 70B
Ultra-fast inference for the reasoning agent — sub-second LLM responses even on complex fault trees.
Knowledge Graph
OWL ontology with 50+ entities provides the structured context agents reason over.