Skip to main content
The product is a pipeline of Cloud Run jobs plus one HTTP API. Collectors write raw files to GCS and some hot-path tables in Postgres. Cleaning and ML workers write typed Postgres tables. The API never scrapes OMIE itself.
1

Collect

Weather, OMIE prices, and ENTSO-E fundamentals land in GCS and Postgres.
2

Clean

Eventarc starts data-cleaning-service when a source writes its manifest.
3

Forecast

After weather cleaning, one forecast job runs per OMIE product.
4

Score

After new OMIE prices, the metric worker scores unscored forecasts.
5

Serve

platform-api reads Postgres for the app, client API, and webhooks.
Jobs are idempotent. A repeated manifest or an already-stored target day exits successfully without double-writing.

​
Triggers

Hourly. Manifest starts weather cleaning, then forecast jobs for all four products.
Every 30 minutes per market (Europe/Madrid). The job self-skips when up to date. Manifest starts pricing cleaning for that market only.
Hourly. Manifest starts fundamentals cleaning.
After OMIE pricing manifests. Scores unscored forecasts against real prices.

​
Pricing weather locations

Forecast models use three fixed locations, not tenant sites.

Madrid

40.4_-3.7

Burgos

42.3_-3.7

South

38.7_-5.1

​
Ownership

Platform team

Collectors, orchestration, Postgres schema (Alembic in platform-api), and the HTTP API.

AI team

Cleaner and forecast/metric model implementations, plus the MLflow registry.