How shreyas30016/forecast-forge works
Forecast Forge AI turns multi-model weather forecasts into adaptive, uncertainty-aware guidance with evaluation and decision provenance. The main path runs from the dashboard through the API and forecast orchestrator to provider data and ensemble results. Separate capabilities cover historical skill evaluation, spatial guidance, extremes, bust detection, probabilistic forecasts, replay, and the trace-grounded copilot. README claims guide the unsampled feature links; sampled code confirms the orchestrator, ensemble engine, replay trace writes, and trace storage behavior.
Main entry points: app.py FastAPI app · service.py Scientific replay · runtime.py Nemotron runtime · page.tsx Dashboard UI
Components 25
User interface
- Dashboard UI
ui/src/app/page.tsx
API surface
- FastAPI app
forecast_forge/api/app.py - Forecast route
forecast_forge/api/routes/forecast.py - Ensemble route
forecast_forge/api/routes/ensemble.py - Evaluation route
forecast_forge/api/routes/evaluation.py - Replay route
forecast_forge/api/routes/replay.py - Copilot route
forecast_forge/api/agent_router.py
Forecast and data
- Forecast orchestrator
forecast_forge/orchestrator/service.py - NWP adapters
forecast_forge/providers/open_meteo/ifs.py - Ensemble engine
forecast_forge/ensemble/engine.py - Forecast validation
forecast_forge/validation/validator.py - Spatial grid service
forecast_forge/spatial/grid_service.py
Evaluation and intelligence
- Historical alignment
forecast_forge/historical/alignment.py - Skill evaluation
forecast_forge/evaluation/scoring.py - Probabilistic guidance
forecast_forge/providers/open_meteo/ensemble.py - Extreme detection
forecast_forge/extremes/service.py - Bust detection
forecast_forge/evaluation/bust/service.py - Scientific replay
forecast_forge/replay/service.py - Trace builder
forecast_forge/trace/builder.py - Trace storage
forecast_forge/trace/storage.py - Nemotron runtime
forecast_forge/agent/runtime.py
Other components
- Forecaster
- Open-Meteo
- ERA5 reference
- Nemotron service
Connections 24
- Forecaster → Dashboard UI: uses
- Dashboard UI → FastAPI app: sends requests
- FastAPI app → Forecast route: dispatches
- FastAPI app → Ensemble route: dispatches
- FastAPI app → Evaluation route: dispatches
- FastAPI app → Replay route: dispatches
- FastAPI app → Copilot route: dispatches
- Forecast route → Forecast orchestrator: requests forecasts
- Forecast orchestrator → NWP adapters: fetches concurrently
- NWP adapters → Open-Meteo: queries
- Ensemble route → Ensemble engine: generates blend
- Ensemble engine → Forecast validation: validates output
- Evaluation route → Skill evaluation: requests scoring
- Skill evaluation → Historical alignment: uses aligned data
- Historical alignment → ERA5 reference: loads reference
- Spatial grid service → Open-Meteo: fetches grid data
- Extreme detection → Probabilistic guidance: requests probabilities
- Bust detection → Probabilistic guidance: requests forecast
- Replay route → Scientific replay: requests timeline
- Scientific replay → Trace builder: builds traces
- Scientific replay → Trace storage: saves traces
- Copilot route → Nemotron runtime: invokes
- Nemotron runtime → Trace storage: retrieves trace context
- Nemotron runtime → Nemotron service: queries