AI-powered chatbot boosts livestock disease management
In an effort to digitalize livestock research and knowledge dissemination, ILRI has developed a new AI-powered chatbot that provides instant, evidence-based answers to animal health questions.
In an effort to digitalize livestock research and knowledge dissemination, ILRI has developed a new AI-powered chatbot that provides instant, evidence-based answers to animal health questions. The tool, accessible at https://animalanswers.ilri.org, was developed by ILRI’s Data and Research Methods Unit (DRMU) and CGIAR’s Digital Transformation Accelerator (DTA) in collaboration with Fahamu AI, a Kenyan startup company.
The chatbot, still in its pilot phase, draws primarily on scientific publications housed in CGSpace, CGIAR’s open-access repository. It is designed to make it easier and quicker for everyone to draw science-based, actionable insights from the CGIAR’s rich body of research on animal health.
Livestock diseases continue to threaten productivity and livelihoods across Africa, with camel, cattle, and small ruminant herders often struggling to access timely, credible animal health advice. Traditional research outputs - scientific papers, reports, and datasets - are not always readily usable by field practitioners who may not have the time for an extensive literature review.
This challenge inspired ILRI’s DRMU team to explore AI technologies that could bridge the gap between academic research and the need for immediate answers to inform real-world applications. According to Jean-Baka Domelevo Entfellner, Head of Data and Research Methods at ILRI, the initiative aims “to make our research outputs truly actionable by letting anyone ask questions in plain language and receive reliable, sourced answers instantly.”
Development and collaboration
The chatbot was developed through a collaboration between DRMU and Famahu AI, with technical support from Alan Orth, DRMU’s systems specialist. The project leveraged ILRI’s existing infrastructure and data repositories, combining AI language models with a curated body of ILRI and CGIAR research on animal health.
Each answer generated by the chatbot is linked to the source materials, called “chunks” that appear as clickable citations. This transparency allows users to trace the evidence behind every response, as illustrated in the screenshot showing a query on tick-borne diseases in Kenyan camels.
“We wanted users to not just trust the answer, but explore the science behind it,” says Orth. “Every response leads back to real scientific publications.”
How it works
The interface is designed to be simple and intuitive. Users can:
- Type a question in the main text box and click Ask.
- Review the chatbot’s response, complete with linked citations.
- Explore the cited text chunks and full references below the answer.
- Provide feedback on the response’s accuracy, completeness, and helpfulness using a built-in star-rating system.
A History feature allows users to revisit past questions and answers both their own and those submitted by others, encouraging community learning and transparency.
Pilot phase and feedback
Currently, the chatbot is in an early pilot phase, with all functionalities operational but still undergoing refinement.
Feedback gathered during a hands-on workshop held on 9 December 2025 at ILRI Campus, Nairobi highlighted both strengths and areas for improvement in the current pilot of the AI-powered chatbot. Participants noted several technical issues that warrant attention, particularly around transparency and accuracy of source attribution. In some cases, answers referenced sources labelled only as “document excerpt,” without clear bibliographic details, which undermined confidence in traceability. Users also observed that relevance scoring of cited document chunks could be overly generous, with multiple sources receiving very high scores despite limited alignment with the query. Additionally, occasional cryptic or unintelligible chunk content appeared in responses, most notably when the system was prompted with deliberately off-topic questions, pointing to the need for improved filtering and error handling.
Beyond bug fixes, workshop participants shared a rich set of suggestions aimed at enhancing usability and flexibility. A recurring request was to move away from a fixed number of five document chunks per response, instead introducing a relevance threshold so that only the most meaningful sources are surfaced. Participants also expressed strong interest in a proper user login system that would allow for private interaction spaces, as well as a fully functional conversation feature to support multi-turn dialogue. Customization emerged as a key theme, with users proposing selectable personas or profiles such as scientist or farmer that could influence tone, verbosity, and depth of responses.
Significance and broader vision
This initiative exemplifies CGIAR’s push to harness digital innovation for agricultural transformation. The chatbot aligns with the Digital Transformation Accelerator’s mission to mainstream AI tools that democratize access to research and enhance decision-making across the agri-food system.
For ILRI, the chatbot marks a milestone in using digital tools to scale livestock research impact. The DRMU team envisions this as a foundation for broader, domain-specific AI tools covering topics such as livestock genetics, nutrition, and climate adaptation. “We’re starting small, but the vision is big,” says Domelevo Entfellner. “With continued support from DTA and our partners, ILRI aims to build a family of knowledge-driven chatbots that can serve farmers, researchers, and policymakers across the livestock sector.”
Looking ahead
Looking ahead, more forward-leaning ideas focused on performance and architectural evolution of the tool. These included offering users explicit control over answer length, exploring the integration of limited large-language-model–generated content alongside corpus-based retrieval, and testing alternative generative AI APIs to optimize speed and accuracy. From a systems perspective, participants also suggested running the retrieval-augmented generation (RAG) pipeline on in-house GPU resources rather than relying solely on CPU-based hosting. Collectively, this feedback provides a clear, user-informed roadmap for strengthening the chatbot’s reliability, responsiveness, and overall impact as it moves beyond the pilot phase.
Users are invited to try out the chatbot - https://animalanswers.ilri.org, explore its references, and share their experiences to help shape the next iteration of ILRI’s digital knowledge tools.