IEEE Humanitarian Technologies

A farming family in Bangladesh receives word that a severe weather event is coming. They have hours, not days, to prepare. Without reliable guidance on what to do — whether to harvest early, move livestock, or protect their seedlings — the decision falls entirely on them. For families with no labor protection, no financial safety net, and no access to timely information, the cost of getting it wrong creates a cascade effect: lost income, disrupted access to healthcare, and a recovery that takes years.

This is the reality that drives the CHP Team and the problem at the heart of our final
finalist spotlight. Dr. Shadi Saleh and Ramzi Halabi have spent years working at this intersection, and they're not building from theory.

At the American University of Beirut's Global Health Institute, Dr. Shadi and Ramzi have spent years at the intersection of climate and health equity. Through their Cli-Health Program, they've seen firsthand how climate shocks affect harvests and the people who tend them. Climate risk, as they put it, is not only an environmental issue. It is a major determinant of health. And the populations most exposed to it are the least equipped to respond.

That framing sets the CHP Team apart. They're building for Bangladesh, but their lens is broader: a world where climate data is abundant, yet the families who need it most are still making high-stakes decisions in the dark.

Their tool would translate complex climate data into hyper-local, actionable guidance grounded in official national guidelines, delivered through lightweight chat interfaces over familiar channels, and designed to work in low-bandwidth environments with simple, step-by-step language. Every response would include clear low-confidence warnings for uncertain forecasts, so that farmers know exactly how much to trust what they're reading.

For the CHP Team, success would mean farmers and agricultural workers are receiving timely guidance, adjusting their practices, and reducing losses from extreme weather events. It would mean the solution has been adopted by regional partners and is reaching more people across more countries. And it would mean farmers trust it because it proved itself when it mattered most.

The winner announcement is coming! Follow the series and stay updated on the Challenge: https://bit.ly/4njszau

"The stakes of getting AI-driven agricultural advice wrong are not abstract—they are life and death."

For Manali, Ayush, and Buvi, those words come from experience.

This week's spotlight turns to Inko, the second of three teams building for the Agriculture use case in Lesotho.

While building a voice bot for farmers in Uttar Pradesh, India, the Inko team encountered something that reframed everything. The farmer su***de rates in the region were alarmingly high—a misidentified crop disease, a wrong remedy, an entire harvest lost.

For a family with no alternative income, that loss can be catastrophic. That experience shaped not just their technical standards, but their understanding of what this work is actually about. As they put it, the urgency is not just technical. It is deeply human.

Each team member brings something essential to that commitment. Manali leads solution architecture and data pipeline design, with deep experience testing for hallucinations and ensuring output quality. Ayush owns the technical pipeline, focusing on scalability, tool selection, and implementation strategies that keep answers accurate and well-grounded. Buvi handles the critical groundwork of structuring and segregating datasets so that when data is vectorized for their RAG pipelines, the AI has minimal room for interpretation or guesswork.

In agriculture, inaccurate advice doesn't produce a bad user experience. It can destroy a harvest and devastate a family.

Inko is working to build a generative AI advisory tool for Lesotho that would give farmers access to timely, reliable guidance—from understanding upcoming weather patterns to planning crop cycles—through low-tech channels like SMS and voice, designed for low-bandwidth environments and limited connectivity. Every response would be grounded in verified, structured data, with human oversight built into the workflow from the start.

"Trust in AI must be earned, not assumed," they write. Their system would be designed with verification at every layer, giving agricultural advisors the ability to validate responses before they reach farmers, and ensuring users always trust their own judgment above an AI-generated answer.

The impact that would make them proud would be the moment a farmer in Lesotho gets the right advice at the right time and makes a decision that protects their harvest, their family, and their future.

If they get this right, it won't just change what farmers know. It will change what becomes possible for the families who depend on them.

Follow the series and stay updated on the Challenge: https://bit.ly/4njszau