Page 849 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 3: AI-Powered Agricultural Contact Center for Farmers Agriculture 4.11: Smart

Organization: Postal & Telecommunications Regulatory Authority of Zimbabwe (POTRAZ)
Country: Zimbabwe

Contact Person: Frank Makeba frankmakeba0@ gmail .com, fmakeba@ cut .ac .zw

1 Use Case Summary Table

Item Details
Category Smart Agriculture

Problem Farmers in Zimbabwe face limited access to timely, localized agricultural advice
Addressed due to a critically low extension officer-to-farmer ratio of 1:2000 well below the
Food Agriculture Organization (FAO) recommended 1:400. Combined with
climate-related shocks and a 30% post harvest loss rate, this lack of support
severely impacts productivity. While over 63% of the population depends on
agriculture, 76% of smallholder farmers live below the poverty line, and 31%
of rural adults are illiterate making traditional digital tools ineffective without
voice support. These challenges fuel food insecurity, poverty, and environ-
mental degradation, threatening national progress toward key Sustainable
Development Goals (SDG) targets.

Key Aspects of An Artificial Intelligence (AI) powered, voice assisted contact center designed
Solution to empower farmers with real time agricultural solutions. Farmers can log their
problems through phone calls, SMS, or a website, and receive automated
voice responses with actionable solutions tailored to their specific needs, local
language support, compatibility with 2G devices, inclusive access via SMS/
voice for illiterate users.
Technology Natural Language Processing (NLP), Speech-to-Text, Voice AI, Local Language
Keywords AI, Predictive Analytics, Chatbot, Climate AI, Internet of Things (IoT) - Ready
Data Availability Data will be available upon request - Agricultural extension datasets from
government and Non Government Organizations (NGO), Localized language
data, Climate Application Program Interfaces (OpenWeatherMap), Crop
disease
Metadata (Type Audio (calls), Text (SMS/web), Visual (satellite & weather data), Structured
of Data) (market prices)
Model Training Fine-tuned Transformer models (NLP & Text To Speech) on local language
and corpora; GPT-2 custom-trained on Shona data (language in Zimbabwe),
Fine-Tuning disease prediction model trained using Convolutional Neural Networks (CNN)
+ local datasets.

813