msresearch-aurora
Choosing the right path
Before diving in, identify what the user is actually trying to do. Three buckets:
-
"I want to actually use Aurora for my problem" (custom variables, regional adaptation, domain-specific predictions) → fine-tuning path. This is the recommended path for serious applications.
-
"I'm exploring / just curious / want to see Aurora work" → inference path (Norway regional example with frontend visualization). Lower commitment, no GPU required.
- Route: docs/about-aurora.md → docs/quick-start-inference.md → docs/customization-guide.md for adapting to a new region.
-
"Not sure / want to see Aurora in action first" → route by audience:
- Non-technical audience (execs, PMs, general scientists): start with the Norway Prototype (docs/quick-start-inference.md) — visual maps over a real region, intuitive at a glance.
- Weather/climate practitioners (familiar with NWP, ERA5, model evaluation): start with the Finetune Exploration Demo (docs/finetune-demo.md) — interactive divergence metrics and batch comparisons that surface fine-tuning behavior. Once they have a use case in mind, steer them toward the fine-tuning path.
If a user mentions a real problem they want Aurora to solve (custom forecasting, air quality, wave heights, regional energy planning, etc.), check the user for their familiarity/comfort with the Aurora model and finetuning in general. If a user is comfortable, steer them toward fine-tuning Aurora. If they want to learn more about Aurora first, direct them to docs/about-aurora.md. If they're familiar with Aurora and/or weather modeling, but they're new to finetuning, send them to docs/about-finetune.md.