Case study / Weather intelligence
Wint
A fully automated service that interprets multiple forecast models and delivers the environmental brief a competitive sailor actually needs.
The problem
The forecast is available. Interpretation is not.
Free weather tools expose maps and numbers but leave the tactical meaning to the sailor. Bespoke meteorologists provide interpretation, but the cost and coordination put them beyond much of club racing, training camps, and smaller championships.
Product idea
Interpretation, not another forecast.
Wint turns ECMWF, GFS, ICON, and ARPEGE output into a clinical race-environment brief: model divergence, stability regime, shift character, upper-air momentum, ranges, and confidence. It describes the environment without prescribing manoeuvres.
Architecture
Python owns facts. The language model owns prose.
- RegattaEngine computes structured weather metrics before any model is called.
- The model receives a compact context object rather than raw forecast logs.
- Generated prose is validated for required sections and length, with local and cloud fallbacks.
- Jinja2 and WeasyPrint combine deterministic metrics with interpretation into a portable PDF.
- A mobile-first PWA controls authentication, access, purchases, and report delivery.
My role
Sailing judgment encoded as product rules.
I designed the briefing around the decisions sailors and coaches make before racing, then shaped the pipeline, validation, report structure, and delivery system. The domain constraint is deliberate: Wint reports the environment and confidence, not tactical commands.
Current state
The pipeline and web foundation are complete.
Data collection, synthesis, PDF generation, persistence, authentication, subscription access, and report delivery are implemented. The next meaningful milestone is beta validation with sailors in a training-camp or championship setting.