Industry: CleanTech / Energy Management
Client: CarbonLnk, Lnk Technologies
Client Location: London, UK (Remote)
Engagement: Freelance
Timeline: February 2026 (1 month)
Type: Multi-Agent AI System
The Problem
Energy management platforms generate large volumes of consumption data. Without automated analysis, detecting unusual patterns, identifying wastage, and giving users actionable advice requires constant manual monitoring. That does not scale.
The client needed an intelligent layer on top of their existing energy data that could watch for problems automatically and deliver personalised guidance to users without human intervention.
The Approach
The solution was designed as a multi-agent system, each agent with a single, focused responsibility. Rather than one monolithic process trying to do everything, three specialised agents work in sequence: retrieve, analyse, advise.
This separation meant each agent could be optimised for its specific task, and the system could be extended or swapped out at any layer without rebuilding the whole pipeline.
What Was Built
A three-agent AI pipeline covering the full energy analysis workflow:
Data Retrieval Agent
Extracts user energy consumption data from the database in real time. Feeds clean, structured data downstream to the analysis layer.
Anomaly Detection Agent
Analyzes consumption patterns to detect spikes, irregularities, and wastage automatically. Flags unusual behaviour without waiting for a human to spot it in a dashboard.
Energy-Saving Advisor Agent
Generates personalised saving recommendations using a RAG-powered knowledge base. Recommendations are grounded in real-world energy-saving practices and tailored to each user's actual consumption profile.
Results
| Area | Impact |
|---|---|
| Anomaly detection | Automated, no manual monitoring required |
| Saving recommendations | Personalised and grounded via RAG knowledge base |
| Analysis speed | Real-time, replacing manual review cycles |
| Energy waste | Reduction in unnecessary consumption through targeted advice |
The Principle
Monitoring energy data manually does not scale. This system replaced a reactive, human-dependent process with a proactive, automated pipeline that catches problems and delivers guidance in real time.
Three focused agents doing one job each well outperform one general system trying to do everything at once.