If you could know before a flood arrives which areas are most at risk, where to prepare first, and which work to prioritise on a limited budget, disaster preparation would be a different exercise. That is the thinking behind Climate Action Copilot, built by two Bangladeshi students, which has reached the final of the AI for Climate Action Award 2026.
Faria Islam studies computer science and engineering at the University of Liberal Arts Bangladesh; Md Habibur Rahman studies microbiology at BRAC University. The award is given jointly by the UNFCCC Technology Executive Committee and the Climate Technology Centre and Network, and exists to recognise AI-based solutions for least developed countries and small island states.
The top five projects will be presented at the AI for Climate Action Forum in Cambodia on 13 and 14 October, where the winner will be announced. Faria Islam travelled on 11 October as team leader. The winning project goes forward to COP31 in Turkey in November.
The question the project asks is the right one
Most climate technology aimed at Bangladesh tries to improve forecasting: better models, earlier warning, finer resolution. That work matters, and it is not where the binding constraint sits.
Bangladesh already receives flood warnings. What it does not have is enough money to act on all of them at once. When a warning covers several districts and the budget covers one, somebody decides which embankment is reinforced, which shelter is stocked and which community is visited first. That decision is usually made quickly, with incomplete information, by people who will be blamed either way.
A tool aimed at prioritisation under a budget constraint is therefore aimed at the actual bottleneck. It is a less glamorous target than prediction and a more useful one.
What would make it work, and what usually kills this kind of project
Two things decide whether a tool like this survives past the award, and neither is the model.
The first is whose desk it sits on. A prioritisation tool is only used if the person who actually allocates the budget — at a union parishad, an upazila office, a disaster management committee — opens it when the decision is being made. Software that produces a better answer than the official process, but sits outside that process, gets ignored politely.
The second is accountability for the ranking. Telling an officer which area to fund first is also telling them which to fund second, and the second area floods too. A tool used in public administration has to be able to show why it ranked as it did, in terms a person can defend in a meeting. A ranking nobody can explain will not be followed, and should not be.
Those are solvable problems, and they are the ones to ask about rather than accuracy.
Worth noting about where this came from
One of the two is a microbiology student. That is not incidental colour.
The useful part of a climate prioritisation tool is knowing what makes one community more vulnerable than another — water, sanitation, disease after standing water, how quickly people can leave. That is domain knowledge, and it is the part a computer science degree does not supply. Pairings like this are what produce tools that work rather than demos that score well.
What winning would and would not do
The prize carries a route to COP31, which is real: a project presented there is seen by the people who fund adaptation, and that is a door most student work never gets near.
It does not, by itself, get the tool into a upazila office. Awards select for the quality of an idea under presentation conditions; deployment selects for whether an organisation will change how it makes decisions. Those are different tests, and the second one is passed by far fewer projects.
So the thing to watch after Cambodia is not the result. It is whether any Bangladeshi agency with a disaster budget agrees to trial the tool on a real allocation, and says afterwards what it changed.
The demand for this kind of training is not in doubt: a government scheme offering 1,600 places drew about a hundred thousand applicants. Whether the schools feeding into it are ready is a separate and less encouraging question.
