An Agentic AI Pipeline for Appliance-Level Energy Anomaly Detection and LLM-Driven Recommendations

Short summary

This paper presents an agentic pipeline combining LSTM forecasting, VAE-based anomaly detection, and LLM reasoning for building energy management. The LangChain system uses dynamic RAG retrieval to maintain accuracy while cutting context from six to three-six sources per event. Evaluation on 16 scenarios achieves 90.4/100 with both cloud and local 7B-parameter model support.

• •Combines LSTM time-series forecasting, per-appliance VAE anomaly detection, and LLM reasoning for actionable facility maintenance insights
• •Dynamic RAG retrieval maintains 90.4/100 accuracy while reducing context requirements by 50%
• •Supports both cloud-based and fully local 7B-parameter LLM backends for flexible deployment

Generated with AI, which can make mistakes.