AI-Powered Chatbot Transforms Farming in Malawi with Multilingual Support

Farmers in Malawi are now benefiting from a groundbreaking AI-powered chatbot that provides vital agricultural advice in multiple languages, according to the NVIDIA Technical Blog. Named UlangiziAI, the chatbot offers on-demand recommendations to help farmers increase yields, combat pests, and control crop diseases.

Multilingual and Multimodal Support

Launched in February, UlangiziAI is accessible through WhatsApp, allowing farmers to send text or voice messages and even photographs for analysis. The chatbot can respond in both English and Chichewa, the latter being the native language of about half of Malawi's 20 million people. This feature is particularly crucial as approximately 30% of the population is illiterate.

Developed for Local Needs

The chatbot was developed by Opportunity International, an NGO with a 21-year presence in Malawi. Recognizing the low smartphone penetration in rural areas, the organization trained local farmer support agents who assist the community using smartphones equipped with the UlangiziAI app. 'Ulangizi' translates to 'guidance' or 'counseling' in Chichewa.

Addressing Structural Challenges

Paul Essene, a key developer behind UlangiziAI, pointed out the systemic challenges Malawian farmers face. With most Malawians living on around $2 per day and 80% residing in rural areas, access to government agricultural advice is inconsistent. Farmers often have to walk long distances, only to find these offices closed. UlangiziAI aims to bridge this gap by providing immediate answers on livestock, plants, fertilizers, and other agricultural concerns.

Technical Framework

The chatbot runs on NVIDIA A100 Tensor Core GPUs in the Azure Cloud, utilizing AI workflow platform Gooey.AI. Given the limited data resources available in Chichewa, Gooey.AI employed innovative techniques to optimize the chatbot's performance. When a voice memo is sent via WhatsApp, it is transcribed and translated into English using advanced language models like Meta’s MMS Large or OpenAI’s Whisper 3. Images of crops or pests are routed to OpenAI’s GPT4o for analysis. The data is then queried against the Malawi Ministry of Agriculture Guide to Agriculture Production Manual, a comprehensive 450-page resource.

Real-Time Solutions

UlangiziAI has significantly reduced the time it takes for farmers to get critical advice—from days to mere seconds. The initial pilot program received over 4,000 queries in both English and Chichewa within three months. Farmers have rated the app highly for its quick and accurate responses.

Future Expansion

Opportunity International plans to expand the chatbot's reach to Kenya, incorporating Swahili and other local languages by late this year or early next year. This expansion aims to provide similar benefits to Kenyan farmers.

For more information, visit the NVIDIA Technical Blog.