Alt Earth

Alt Earth: AI for Earth Sciences

Humanity can now see across astonishing scales. We can image stars in galaxies billions of light-years away, and reconstruct proteins atom by atom. While we master the cosmic & the microscopic, we struggle with the planetary.

We understand remarkably little about the Earth’s natural systems: its rivers traversing continents, its geological structures hidden beneath our feet, trenches deep in the ocean, or even fungi weaving through soil. These natural systems sustain the building blocks of our civilisation—agriculture, energy, the climate that sustains life, and the natural beauty that makes us marvel.

Our planet’s messy, complex natural systems cannot be learned from the internet alone. Understanding them requires intelligence grounded in physical observation. Nature is not merely a static dataset; it is perhaps the richest learning environment we have. We can now build models that remain in contact with the physical world: observing, forming hypotheses, testing them against reality, and recalibrating. Weather forecasting already shows what this can look like. Satellites, weather stations, radar, ocean buoys, aircraft and atmospheric sensors continuously generate new observations, which are assimilated into models that constantly improve against what happens next.

The same principle applies across other fields in the Earth Sciences. Just as computational biology is transforming drug discovery and computational chemistry is making breakthroughs in discovering new materials, we believe models capable of representing the state and dynamics of the Earth can transform how humanity discovers, develops and manages its natural resources.

Alt Earth is building AI for Earth Sciences. The next frontier for AI is intelligence capable of understanding the physical systems of our planet.

Here’s how:

  • Bringing together interdisciplinary teams of Earth scientists, ML engineers, applied mathematicians & on-ground operators
  • Building lab infrastructure at a generational scale
  • Deploying sensors widely to generate high-quality labelled data
  • Training models by deploying them in the physical world

We began with carbon removal. Enhanced Rock Weathering (ERW) required understanding how rock, soil, water and climate interact across real landscapes. So we built field operations in Darjeeling and laboratory infrastructure capable of measuring them. Thousands of samples gave us proprietary data on soil chemistry, mineral composition, weathering and carbon movement. We used those observations to build models, used the models to guide where and how we intervened, and constantly measured the landscape to trace the final outcomes. That’s how we have removed ~10,000 t of CO₂ from the atmosphere to date.

Our next frontier is the subsurface. It underpins the minerals and energy systems that sustain modern industry, provides reservoirs for carbon and waste storage, and hosts many of the geological processes behind natural hazards. Yet we still cannot reconstruct the hidden structures of the subsurface from sparse observations. This is an inverse problem at planetary scale.

Solving it would open an entirely new frontier for humanity, allowing us to understand rock bodies, faults, mineral systems, fluids, heat, and the pathways through which they move. By applying AI to build three-dimensional models of the subsurface, we can unlock an age of abundance—changing how we discover minerals for the green transition, develop geothermal resources, store carbon and assess geological risk.

Alt Earth combines machine learning with geophysics, geochemistry, remote sensing, laboratory analysis and field experimentation to build models grounded in physical observation. Our aim is to reconstruct the subsurface in 3D, together with the geological processes and history that shaped it.

Our scientific mission is to develop Natural General Intelligence—AI that learns from the Earth itself.

When scientists & engineers, theoreticians & experimentalists work together, science moves forward. LIGO built on Einstein’s theory of general relativity and gave humanity new tools to observe the Universe. Sonar, magnetometry & seismological innovation confirmed Wegener’s ideas, giving us plate tectonics, a new way to understand the planet. The 2026 Nobel Prize in Physics recognised IceCube, which turned a cubic kilometre of Antarctic ice into a telescope for neutrinos. By looking down, we have made new discoveries about our vast cosmos. The subsurface demands the same spirit: new theories, new instruments, new models and sustained experimentation brought together to make an inaccessible world observable.

This is a Journey to the Centre of the Earth.

Our Team

Sourav Ganguly, Sambuddha Misra, Pawan Bharadwaj, Matthew Morris-Cave, Adrian Marin Mag, Aalok Thakkar, Abhishek Kumar, Sanjutha Indrajit, Leander Cibi, Miaozhou Tang, Arya Tilak, Sanket Bajad, Meesha Chotai, Akhtar Jahan, Usman Khan, Sankhadip Poyra, Shrey Agarwal, Eeshana Aurora, Kavya Kurian, Sparsh Agarwal

Sourav Ganguly

Geology & Isotope Geochemistry; postdoc, Weizmann Institute

Carbon Subsurface

Head of Data

You will own the data from which we infer the subsurface, including what we acquire, how we represent it, and whether a model can rely on it. The central problem is heterogeneity. Gravity, magnetics, electromagnetics, seismic, geochemical assays, drill-core scans, geological maps and satellite spectra each measure a different property, at a different scale, through different physics and with different errors. We want to model these properties of the measurements, not clean them away. Ground truth is also scarce: only a few deposits are well documented.

What your first year looks like

You have:

Nice to have:

ML Engineer

You will build the models that learn from large volumes of Earth observation data and turn indirect measurements into calibrated predictions about the subsurface. We are hiring two engineers. One will focus on representation learning and the other on generative and probabilistic inference.

There are two problems here. The first is learning from data with very few labels. We have terabytes (and in a few regions, petabytes) of satellite imagery, hyperspectral scans, airborne geophysics, drill-core imagery and multi-element assays, but only a few hundred known deposits. Those labels are positive-only and spatially biased, and the inputs shift between sensors, regions and acquisition conditions.

The second problem is inference. The subsurface is observed only through physical forward models, so the inverse problem is ill-posed and non-unique. The task is to learn priors over 3D geology and to build fast surrogates for the physics, so that we can compute posteriors that stay calibrated.

What your first year looks like

Track A: Representation learning

Track B: Generative and probabilistic inference

Both tracks

You have:

Nice to have:

Applied Mathematician

You will develop the probabilistic foundations of subsurface inference. That includes making sure every model returns calibrated uncertainty, and that every new measurement is optimally chosen.

The central problem is to infer high-dimensional fields from data that constrain them only indirectly, through physical models that are expensive to evaluate. Our unknowns are 3D distributions of physical and geological properties, often with millions of parameters, and the forward models that link them to observations are only approximately correct. The data is compositional, censored and non-Gaussian, and each new measurement is expensive. Deciding what to measure next is therefore pivotal.

Our team has strength in adjoint-based geophysical inversion. This role adds rigorous probability, spatial statistics and decision theory. You will work alongside inversion researchers, ML engineers and geoscientists.

What your first year looks like

You have:

Nice to have: