Prototype · energy forecasting
Windline
Wind-power forecasts that account for what weather information was actually available at the time.
Windline was developed at HackAlem AI in Astana on 23 September 2026. The prototype forecasts hourly normalized active power for two wind turbines over a 24–48-hour horizon.
01 / THE QUESTION
Can a forecasting workflow avoid learning from weather data that would not have been available at the historical forecast time?
HackAlem AI · Astana, Kazakhstan · 23 September 2026.
- 01Available weather & SCADA
- 02Power forecast
- 03Publish / shadow / reject
A prototype forecasting and validation workflow, not a measured performance curve.
02 / THE WORK
My contribution
Worked on a vintage-aware workflow using historical SCADA data, weather forecasts, and LightGBM.
Contributed to checks of forecast inputs and outputs and publish, shadow, or reject decisions.
Integrated forecasts into a FastAPI and React web workflow with history, provenance, and automation logs.
Methods & focus
- LightGBM
- Weather-vintage checks
- FastAPI
- React
- Forecast provenance
03 / EVIDENCE & LIMITS
What the work supports
A hackathon prototype connecting prediction with validation and recorded decision-making.