A lot is being written about AI agents. In most cases, a toolkit for one agent will be sufficient. However, sometimes it will be necessary to create multiple agents, with specialized skills and they should work together.
I created a demand forecasting system that had three agents: a forecast agent, a risk analysis agent and a narrative agent. In this blog post, I will outline the lessons I learned about the complexity and trade-offs of creating a multi-agent system.
What is the Architecture of the Forecast Agent?
The demand forecasting system is a system that predicts how much inventory a retail business will need. A single LLM call is not sufficient to get a good forecast, as forecasting has three distinct skill sets:
- Statistical Modelling (e.g. time-series analysis, seasonal decomposition, trend detection) ...