Eisengrau

Eisengrau (built as GridPulse) brings GB wind forecasts, demand forecasts and imbalance settlement prices together with forecasting and statistical analysis. It has a daily Overview, a dockable Terminal of linked charts and tables, and an AI agent that can build an investigation from a plain-language question. We built it in the 25-hour OxHack 2026 window.

What I built

• The API. A FastAPI backend over Tiger Data (Postgres with TimescaleDB) that serves chart-ready series, algorithm parameters, stored runs and the datasets behind each result.
• The algorithm runner. Adapters for the team’s forecasting methods, with every completed run stored alongside its version, settings and coverage, so the frontend can compare results without recomputing them.
• Backtests and statistics. Rolling-origin forecast backtests with Diebold–Mariano tests, plus price decomposition and stationarity analysis.
• Making it fast. Backtest folds fitted in a process pool and memoised between runs, and latest forecasts materialised so all series load in one query.
• Shipping it. Docker images and GitHub Actions that tested and deployed separate development and production environments to one Vultr VPS on every push, plus a plain-language status page.

What I’d take forward

Storing analysis results with their provenance made the rest of the product easier: the frontend, the agent and the status page all read the same records, so any number on screen can be traced back to its source data.