ProjectsB3
Nuboria
Is this sunrise or sunset worth going out for, when, and from where?
- Role
- Creator / Product Lead — AI-assisted development
- Since
- August 2026
- Status
- In development. Not public yet.
SummaryAerial view
Nuboria is a weather and geospatial product that helps decide whether a sunrise or sunset is worth watching, at what time and from which place. It started in August 2026 with a clear scope: Spain first.
- Date
- 7 September 2026
- Scope
- OpenStreetMap places, release es-geofabrik-20260907-v1.
- Source
- OpenStreetMap
© OpenStreetMap contributors. Weather data: Open-Meteo.
Is this sunset worth it?
Illustrative simulationSynthetic dataMove the controls. The recommendation explains itself, and the score never pretends to be a probability.
This demo needs JavaScript. The case below reads the same without it.
CaseOn foot
Problem
A weather forecast on its own does not answer the question. Whether the sky will be worth it also depends on the sun's path, the horizon, the terrain and the exact place you stand.
Decisions
A deterministic score before any model
BusinessEvery recommendation can be explained rule by rule. If the product says go, it also says why.
TechnicalSolar, Terrain and Visibility layers feed a deterministic score. It is rule-based: there is no trained model.
Confidence is a separate number
BusinessHow promising the sky looks and how much the data can be trusted are two different questions, so they get two different answers.
TechnicalConfidence is computed apart from the event score, on versioned weather data from Open-Meteo.
Prepare the evidence before the learning
BusinessThe product is designed to collect feedback first. Any future learning has to rest on real observations, not on the heuristic grading itself.
TechnicalFeedback and dataset contracts are defined in Supabase with PostGIS. The learning layer stays on the roadmap.
How I checked it
A controlled test with Open-Meteo data gave satisfactory results.
What I learned
An honest heuristic is more useful than a number that looks like a probability. Separating score and confidence forced me to say what the product knows and what it only assumes.
Role
Problem, scope, product and architecture decisions, scoring rules, priorities and the limits of what it may claim. Implementation with Codex, UI with Lovable, and support from ChatGPT.
EvidenceBelow the surface
Architecture
- Data
- Service
- Person
The same flow, step by step
- OpenStreetMap places and Open-Meteo forecasts enter as versioned data.
- Solar, Terrain and Visibility layers describe the scene for a place and a time.
- Deterministic rules turn the scene into an event score.
- Data confidence is calculated separately.
- The recommendation shows both, with the reasons.
Product plates
Plate 1 of 5The atmospheric decisionEditorial compositionSynthetic data Plate 2 of 5Sun and horizonEditorial compositionSynthetic data Plate 3 of 5The recommendationEditorial compositionSynthetic data Plate 4 of 5The index, explainedEditorial compositionSynthetic data Plate 5 of 5Evidence for future learningEditorial compositionSynthetic data
Figures and where they come from
- Date
- 7 September 2026
- Scope
- OpenStreetMap places, release es-geofabrik-20260907-v1.
- Source
- OpenStreetMap
© OpenStreetMap contributors. Weather data: Open-Meteo.
Stack
Page last revised: