From hackathon to global recognition: How Bangladesh's youth-led AI innovation is reshaping cyclone preparedness

Every early warning system carries an unspoken assumption about who it is built for and who it is built by. Too often in disaster-prone countries, that answer comes from somewhere else, a model designed elsewhere, in another context, then adapted for local use. Bangladesh's coastline, and the cyclones that regularly threaten it, called for something built the other way around: by local youth who understand the risk, in the language communities speak, governed by the institution responsible for keeping them safe.

This is the premise Start Bangladesh Hub set out to test when it launched the FOREWARN Disaster Hackathon: that young innovators, given real access to experts, humanitarian actors, and government partners, can build tools that hold up under global scrutiny, not just local goodwill. And the milestone we're sharing now is evidence that the premise held. 

The pilot tool "AI-Driven Impact-Based Cyclone Guideline Generation for Bangladesh," born out of the hackathon, has been featured as a case study in the  Leveraging AI to Enhance Multi-Hazard Early Warning Systems (MHEWS) report, published under the Early Warnings For All initiative by the United Nations Office for Disaster Risk Reduction (UNDRR), the World Meteorological Organization (WMO), the International Telecommunication Union (ITU), and the International Federation of Red Cross and Red Crescent Societies (IFRC). 

From youth idea to a national tool 

The story behind this milestone began with FOREWARN Bangladesh's Disaster Hackathon 2.0, held between April and September 2024. The hackathon brought together university students and early career professionals from across the country. They were to design locally relevant solutions to humanitarian and disaster risk management, working alongside multidisciplinary experts and technical partners. 

Out of that process emerged Team SIDR, led by Bangladesh University of Engineering and Technology (BUET) student Md. Abrar Faiaz, alongside Alphy Shaharin and Nowshin Nawar. Their winning concept combined artificial intelligence driven cyclone impact-based with integration of LLM to generate location specific guidelines, directly tackling one of the most persistent challenges in cyclone anticipatory action: how to translate technical forecasts into precise, impact-based guidelines that decision makers and communities can readily understand and act on. 

That idea did not remain on paper. It has since progressed into the piloting phase of an “AI-Driven Cyclone Impact-Based Guideline Generation for Bangladesh” tool, has been developed through collaboration between the Bangladesh Meteorological Department (BMD), Team SIDR, and the FOREWARN Bangladesh team. 

How the tool works 

The “AI-Driven Cyclone Impact-Based Guideline Generation for Bangladesh” uses a Retrieval-Augmented Generation (RAG) framework. It explores how generative AI can translate real-time weather forecasts into localized, Bangla-language, impact-based guidelines to support early action in coastal Bangladesh. The tool has been trained across different intensity classes based on wind speed, storm surge, and rainfall as well as detailed anticipatory actions corresponding to different impact scenarios and multiple sectors. 

A cyclone damage classifier is used to estimate probable impacts based on forecast conditions. Moreover, multi-layer risk maps help visualize spatial patterns of risk. They also identify areas that may require earlier or intensified preparedness. The retrieval layer selects relevant local exposure and vulnerability information.  The generation model then uses this information to produce context-specific guidelines for appropriate anticipatory action. 

The aim is to close the gap between what a forecast says and what decision-makers and local communities in cyclone-prone areas need to know and do, before a cyclone impacts them.  

The tool is not meant to replace the judgement of meteorological, humanitarian, or local authorities. It is designed as a decision-support system that can help translate technical weather information into clearer, more locally relevant guidelines. 

Currently, the system is still being developed and tested. Early results suggest that localized advisories can improve the clarity and operational relevance of early-action information. They can also help identify potentially high-risk areas where anticipatory action could be prioritized. At the same time, the pilot highlights the importance of high-quality local data, culturally appropriate communication, and continued validation with local experts and institutions. 

Why this matters beyond Bangladesh 

The significance of the pilot tool lies not only in its use of AI, but in how the technology is being developed. Rather than taking an externally developed early warning solution and adapting it to Bangladesh, the pilot is being shaped around Bangladesh's own hazards, data, language, institutional context, and operational needs. It brings together members of the FOREWARN Disaster Hackathon Team, technical experts, humanitarian actors, and the Experts from Storm Warning Centre of the Bangladesh Meteorological Department to explore how a locally rooted solutions can be developed and put to practical use. 

This distinction matters. Early warning systems across much of the Global South have historically been influenced by a largely top-down, North-to-South model, with technical expertise, financing, and decision-making often concentrated outside the countries and communities where systems are ultimately expected to operate. While external support can be valuable, such a model can also create long-term dependencies. Especially, as the knowledge, infrastructure, and institutional capacity required to maintain a system do not remain locally embedded. 

FOREWARN Bangladesh is exploring an alternative approach. It is a Bangladesh-led effort to develop a tool that reflects the country’s cyclone risks and local information needs. The guidelines are generated in Bangla, and the tool is being developed in collaboration with national meteorological experts.  

Building an ecosystem for youth-led innovation 

The objective is not to simply build another AI application, but to explore what locally led innovation can look like when young people, technical experts, humanitarian actors, and national institutions are involved in shaping a solution from the outset. At Start Bangladesh Hub, we see the FOREWARN Bangladesh pilot as an early example of the ecosystem we are working to build, one that connects young innovators with the expertise, partnerships, and institutional pathways needed to turn locally identified challenges into practical solutions. 

It demonstrates the potential of intersectoral collaboration. The FOREWARN Disaster Hackathon brought together young innovators with technical expertise and humanitarian networks, creating a pathway through which an initial idea could be developed as a practical tool with institutional partners. The longer-term ambition is to understand how such tools can move responsibly towards adoption without losing local ownership or community relevance. 

However, the tool should not yet be presented as a completed or fully operational early-warning system. The technology remains under development, and further testing, validation, data improvement, institutional processes, and responsible deployment will be required. Its value at this stage lies in demonstrating a promising pathway from locally identified problem to youth-led innovation, and then towards technical development and government engagement. 

Its inclusion in the MHEWS report places the pilot within a wider global discussion about how AI may strengthen multi-hazard early-warning systems. The report emphasizes that AI is an enabling technology whose effectiveness depends on strong institutions, governance, human oversight, and people-centred and equity-driven design. The FOREWARN Bangladesh pilot reflects these principles by positioning AI as a decision-support capability that complements, rather than replaces, meteorological expertise, humanitarian judgement, and local knowledge.