Decide before the record breaks.
Kairom forecasts droughts, heat and floods past anything on record, and shows every response your agencies can take together: who has to agree, and what each week of waiting costs.
For water agencies, civil protection and utilities. Built on public data; no agency uses it yet.
Replayed on real crises
Seven crises, four continents.
Each replay feeds the engine only what was known at the time, then scores its plan against what really happened.
The problem
Extremes are breaking the records forecasts learn from. And the response is split across agencies that each hold one piece of it.
Forecasts trained on the past
When a drought goes past the record, models built on history keep expecting the usual, at the moment it matters most.
No one sees the whole set
The water agency, the state, the port, the city: each controls some options. The best response often needs several at once.
Waiting has no price
Delaying a decree, a contract or a transfer by a month feels free. It isn't, and nobody puts a number on it today.
How it works
From a forecast to an agreed response.
Every week Kairom reads your live data, updates the chances, and re-plans the response across every agency involved.
Forecast past the record
A model trained on physics-generated climate scenarios, so it has seen conditions the record hasn't. Below: the emergency risk it showed on each replay date, using only the data known that day.
Alert on your plan's lines
Chances for the stages of your own drought plan, never lower than past years alone would show.
Combine across agencies
Every action every body could take, with costs, lead times and physical links. Lines show where one action needs another body's agreement or assets.
Hover or tap an agency to see what links it to the others.
Ask in plain language
An AI layer that can only run the engine and show its results. Every number in an answer comes from a run.
Replay
What Kairom would have said on the day.
Drag through Catalonia's 2022–24 drought, or press play. Each step uses only the data available that day.
What happened next
The full plan on five decision dates:
Ask
What-ifs, answered by the engine.
Real answers from the prototype, shortened. If a question is beyond the model, it says so instead of guessing.
Evidence
Tested on what happened.
Replays use only the data available on each date and are scored on the outcome. The forecasting test was registered before we ran it.
From perfect foresight
Plans made with Kairom's forecasts, over 28 dates in Catalonia 2022–24. Plans using past years alone: 0.41%.
Of warning
On 29 Aug 2023 Kairom put an emergency within 26 weeks at 43%. Storage crossed the line in early February 2024.
Less forecast error past the record
Our physics-trained model combined with Chronos-2, on held-out record-breaking periods. A control model added 1%.
What we don't claim yet
Better you hear it here than in a pilot.
- In the 2022–23 winter every model we tried, ours included, expected the usual refill. Alerts now never show less risk than past years alone would.
- Past the record, uncertainty bands are still too narrow for every model tested: they cover about 60% of outcomes where they should cover 80%.
- Forecasting how a decision changes the outcome is not supported yet. Kairom values decisions against forecasts; it doesn't forecast their effect.
- Combined actions gain only where a physical or legal link exists. In most cases tested, the gains simply add up.
Questions
Before you ask.
Is this a weather forecast?
No. Kairom forecasts what your plan runs on, such as reservoir storage, river levels and heat, weeks to months ahead, and turns that into decisions: what each costs and who has to agree.
Does Kairom make decisions for us?
No. It shows options, chances and costs; people decide. Each recommendation names the body with the authority to act and the ones that must agree.
What data do you need from us?
We start from public data: reservoir levels, river gauges, temperature and your published drought or emergency plan. Your own inventories (plants, contracts, lead times) make it sharper, and you choose what to share.
Where does it run?
The engine runs on Google Cloud in the EU (Belgium). The AI layer uses Google's Gemini models; it can be confined to EU processing, or switched to another provider, to meet your rules.
Which hazards does it cover?
Drought and water supply first, then heat and river floods. The decision engine has also been replayed on Hurricane Harvey, the 2022 Pakistan floods, the Rhine low-water crisis, the Horn of Africa drought, three Bay of Bengal cyclones and the French 2022 heatwave.
Can we see it with our own region?
That's what the waiting list is for. Tell us your region and the decision you find hardest, and we'll tell you honestly whether Kairom can help with it yet.
Waiting list
Bring Kairom to your region.
We're opening Kairom to a small number of water agencies, civil-protection services and utilities. A person reads every request.
- We reply within two weeks, to the address you give.
- No newsletter: we only write about your request.
- Your details stay with us and are never shared.
You're on the waiting list.
We'll reply within two weeks.