Forecast

Demand forecasting typically requires data scientists and weeks of work. Small businesses can't afford that but still need to avoid stockouts and overstock.

Built a multi-agent system using CrewAI. Three specialized agents collaborate to produce actionable forecasts: a Forecaster analyzes the data, a Risk Analyst flags anomalies and a Narrator writes executive summaries.

1. Agent orchestration: each agent's output feeds the next, so one failure cascades down the chain
2. Confidence band calibration across different product categories
3. Narrative generation that's specific enough to be actionable, not generic summaries
4. Handling sparse data for new products with limited history