We are thrilled to announce the launch of Ask ReliefWeb – a tool powered by Artificial Intelligence (AI), designed to enhance the user experience on ReliefWeb by allowing users to ask questions about individual reports through a chatbot-like interface. The chatbot responds with content exclusively from the report in question, ensuring the consistency of the information provided.
To access Ask ReliefWeb, users need to create an account and/or login to their ReliefWeb account, browse to any report, and click the chat button on the bottom right of the screen. Specific questions have a higher chance of better results as we have enforced a strong matching requirement – helping to avoid mistakes or hallucinations – as we err on the cautious side.
Although this is only a first step in leveraging AI on ReliefWeb, we believe that it will already enhance user experience by enabling readers to quickly get information they want by asking simple questions, such as:
- What is the situation in Location A?,
- What are the most acute needs in Location B?, or
- How many people need health assistance?
As this is a very new service, we are keen to hear your feedback. After each question, you are welcome to use the thumbs-up, thumbs-down or detailed feedback form to let us know how this AI services is working.
The technical bits
The Ask ReliefWeb solution has adopted the Retrieval-Augmented Generation (RAG) approach thereby ensuring that answers will only come from content stored on ReliefWeb or inside the Large Language Model (LLM) itself. This solution greatly reduces any risk of mistakes or hallucinations.
For the implementation we are using Amazon’s Bedrock Generative AI service and Titan Foundational Model, Drupal, Elasticsearch, MariaDB, and MuPDF.
ReliefWeb’s AI Future
At ReliefWeb, we embrace innovation to enhance our operational efficiency and effectiveness. We are actively exploring AI opportunities and will be carefully testing and deploying solutions for our team and users over time. Stay tuned for future posts about our AI-powered initiatives.