ReliefWeb and HAQM distribute life-saving insights faster with generative AI

When natural disasters and conflicts create survival situations, the rapid response of the world’s hundreds of humanitarian organizations saves lives. Those organizations can save even more lives when they acquire information in pace with fast-evolving situations on the ground. HAQM is investing $25,000 in AWS credits and providing hands-on engineering support to help the United Nations’ emergency reporting service, ReliefWeb, get up-to-date crisis information to everyone who needs it—and make sure that no situation is forgotten.

For almost three decades, ReliefWeb has provided real-time intelligence by digitally compiling the reports of aid workers working in emergency zones. Launched by the United Nations Office for the Coordination of Humanitarian Affairs in 1996, it helped to revolutionize delivering aid by making the information previously buried in faxed reports available to all humanitarian organizations online.

ReliefWeb’s information management teams are working with HAQM to transform what those organizations can do with the content it gathers. By unlocking the potential of generative AI and large language models (LLMs), they can make life-saving insights more searchable and more actionable. The original launch of ReliefWeb reduced the time it takes to issue situation reports from 2 weeks to 24 hours. Generative AI can cut the time it takes for humanitarian organizations to get the information they need from hours to seconds.

“There is no reason to reinvent the wheel when the materials are out there,” explains Sandra Dacosta, who works for Norwegian People’s Aid in Iraq. “ReliefWeb saves many hours of my time so I can focus on other critical matters. Other websites are targeting specific aid workers. ReliefWeb is targeting everyone. It is diverse and inclusive.”

Giving voice to the people on the ground—at every emergency

ReliefWeb has more than 18.2 million users, over 14 million of which actively download its emergency-situation reports. In 2023, 50,950 such reports were added to the site, allowing teams in regions such as Yemen, Ethiopia, Haiti, and Myanmar to get the word out about their emergencies—even when the eyes of the world’s media are focused on other higher-profile issues. The information they share helps make sure that the right type of humanitarian support arrives in the right places: that refugee camps are established in locations that refugees can safely reach, or that ambulances are deployed in areas with surviving hospital infrastructure to support them. Following natural disasters, reports such as these can help determine which forms of communications are still working and which roads are still accessible.

As weather alerts were issued about the likely impact of El Niño rains on Somalia, Liston Mwabi, the assessment officer at REACH Initiative, needed to quickly develop a strategy for using cash assistance to alleviate the impact. The challenge was that there was no information available on the likely impact of this type of aid in his area.

“I turned to ReliefWeb to explore how other countries had approached assessments of this kind,” Mwabi explained. “I looked at similar situations in Ukraine, Haiti, and Afghanistan. The Somali Cash Consortium used the reports to help guide preparedness, and I believe our own reports can serve as a benchmark for similar assessments in other contexts.”

“No emergencies are ever forgotten to us,” said Andrew Alspach, chief of the Information Management Branch (IMB) at UNOCHA. “ReliefWeb exists to give a platform to those monitoring these situations and pool information in a way that makes a difference. Getting the attention of politicians, the media, and aid donors is often challenging, and more so as conflict spreads and the impacts of climate change are felt more widely. That’s why it’s so important for ReliefWeb to keep innovating to make our information accessible—and push it out to everyone who needs it.”

Instant insights in natural language with HAQM Titan

ReliefWeb and HAQM developed Ask ReliefWeb, an AI-powered chat-based application, which lets users to ask questions about the content of any report on the site and get instant answers with the information they need. Built using the HAQM Titan set of foundation models (FMs) available in HAQM Bedrock, Ask ReliefWeb responds to natural language queries. It does so through a Retrieval Augmented Generation (RAG) approach in which answers draw solely on the content of ReliefWeb reports and carefully curated HAQM Titan LLMs. This maintains the crucial integrity of ReliefWeb content while deploying generative AI to surface life-saving details faster.

“The launch of Ask ReliefWeb is really the first step in an iterative innovation process that we’ve been developing with HAQM,” said Alspach. “We’ll be able to analyze the types of questions that our users ask and the performance of our models in answering them accurately. Once we’ve got that insight, we can explore how viable it is for the chatbot to answer a broader range of questions using information from several different reports at once.”

Preparing data to make a difference

It’s not just the Ask ReliefWeb chat-based assistant that lets users access ReliefWeb insights faster. HAQM has provided credits and dedicated engineering teams to work with ReliefWeb on vectorizing the database it stores on AWS. This involves using machine learning (ML) to categorize unstructured data, capture its meaning and significance, and make it available to humanitarian organizations as an API.

“This makes a really huge difference to the type of organizations we work with, which don’t have their own budgets for storing and processing vast amounts of data,” said Alspach. “Having ReliefWeb data available as an API means that they can write their own generative AI code to analyze and interpret it, enabling them to do even more with the information we supply.”

Besides getting information out quickly to those who need it, generative AI can also help transform the speed at which ReliefWeb compiles it. Alspach’s team explores the potential of LLMs to analyze scanned PDFs and create instant summaries of the key points for more rapid publishing. At the same time, they work with AWS to develop models that categorize and tag the tens of thousands of humanitarian jobs and training opportunities posted on ReliefWeb, connecting those with relevant skills to the situations that need them.

“HAQM support has been really valuable in helping us to develop proofs of concept like this and tackle the different barriers that we face,” said Alspach. “We can start with testing the effectiveness of HAQM Titan LLMs for categorizing job submissions, and if that’s successful, we can look at categorizing our reports the same way, helping organizations to submit and distribute them even faster. It’s a continuous process that comes down to unlocking the potential of AI for powering humanitarian analysis at scale.”

Visit ReliefWeb and AWS to learn more about how AI is used to respond to disasters and save lives.