Internal forecasting and trading copilot for a Tier-2 UK energy retailer, blending classical models with LLM-driven scenario narratives.
Kestrel’s trading desk needed scenario narratives — not just numbers — to brief commercial teams under tight deadlines. We paired a classical forecasting model with an LLM layer that reads the model output and writes the trader’s morning note, plus a chat surface for ad-hoc questions against the historical data warehouse.
The system has been in production since Q3 2025. We continue to operate the eval suite under a small monthly retainer.