meneltir°
We are Meneltir: a research-driven team building planetary-scale weather intelligence powered by novel, AI-native satellites and state-of-the-art machine learning. Our goal is simple and testable—to turn raw, global signals into timely, trustworthy guidance that improves decisions in shipping, energy, insurance, agriculture, and public safety. We work like a modern research group: co-designing sensing and models, publishing methods and benchmarks, and validating results with partners who run real operations. The output is not a dashboard for show; it is an evidence-backed stream of forecasts and risk indicators that people can use with confidence.
What We Do
Meneltir unifies observation, modelling, and decision support. Our novel satellites are engineered to observe what matters for high-impact weather and to run inference on-orbit, reducing latency and bandwidth while increasing reliability. On the ground, we fuse these observations with authoritative external datasets and probabilistic, spatio-temporal ML to produce consistent global fields—precipitation structure, temperature and heat indices, wind and gust risk, and short-range evolution. Everything is delivered through stable APIs, SDKs, and applications that integrate with routing, grid, claims, and emergency-management systems. We emphasise calibration and interpretability throughout: models are scored on open benchmarks, uncertainty is quantified, and every alert can be traced to the evidence that generated it.
Why It Matters
Weather forecasting is undergoing a revolution. For decades, numerical models have dominated—vast simulations running on some of the largest supercomputers ever built. But they are slow, expensive, and limited by the very gaps in data that Meneltir is designed to fill. A new wave of machine learning models, from Aardvark to Aurora, has shown that data-driven methods can rival and even surpass physics-based forecasts.Meneltir is uniquely positioned at the intersection of these two revolutions: we are not only building an orbital sensor network that closes the data gap, but also designing AI systems that learn directly from it. By fusing harmonized multi-sensor observations with end-to-end, physics-informed learning—self-supervised pretraining on radiances, differentiable data assimilation, and neural-operator cores—we deliver rapid, calibrated forecasts that sharpen as the constellation densifies.This is not just an upgrade. It is a step change in how humanity sees the skies.
Join Us
This team is research-engineering, not research-theater. We write the forward models we need, the flight code we can sleep on, and the ablations that close the gap between a paper figure and an operational timeseries. We value slope over pedigree, clear writing over cleverness, and reproducibility over conference applause. You’ll see thermal budgets, RF chains, compiler flags, and reliability plots in the same review. You’ll ship something that orbits and something that publishes, and both will survive contact with reality.If you want a lab in orbit, if you care about microphysical truth and operational impact in the same breath, if you believe the right abstraction boundary between physics and learning is built—not debated—come build it here. Professors with groups to co-lead retrievals and DA; postdocs who want to turn a method into an instrument; engineers who know why an interrupt storm ruins a pass; statisticians who wake up thinking about calibration curves—this is your invitation. The age of blind spots ends when we decide to measure them.