The problem the AI tool aims to solve is inadequate extension services in Tanzania.

Extension services in Tanzania are subsidized by government extension officers. Due to limited human resources and geographical barriers, only 7% of crop-growing households reported to receiving these services. This has led farmers to struggle to access timely and accurate information necessary for effective farming decisions.

• Low horticultural productivity.
• The horticulture sub-sector may fail to meet its intended growth in exports of USD 2billion by 2030.
• Smallholder farmers may continue to rely on ineffective farming techniques.

The AI tool addresses the specific issue of inadequate access to timely and accurate horticultural information.

Low adoption of modern farming techniques, high post-harvest losses, improper use of inputs, and minimal access to extension services.

Extension services, farmer training programs, radio and television programs, NGO-led initiatives, and printed media such as newspapers.

Using natural language processing techniques to provide responses to farmer queries in Swahili language. By providing farmers with timely information to enhance decision-making. Complementing traditional extension services by reducing the workload of extension officers, allowing them to focus on complex cases.

It is estimated that less than 10% of smallholder farmers in Sub-Saharan Africa have access to regular extension services.

Only 7% of crop-growing households in Tanzania receive agricultural extension services.

Shortage of extension officers. The ratio of extension officer to smallholder farmers is 1:2000.

• Limited reach of traditional extension services.
• Population growth and food demand.
• Economic importance of the horticulture sub-sector.
• Post-harvest losses.

The purpose of the AI tool is to improve access to horticultural extension services for smallholder farmers in Tanzania.

The tool was developed by researchers from AfriAI Lab.

The tool uses natural language processing techniques (machine reading comprehension) to understand the user queries and respond.

Horticulture farming guides in Swahili language were collected and used to create a dataset for training the AI tool.

The collected horticulture farming guides were validated by extension officers based on their accessibility, relevance and comprehensiveness.

The dataset was annotated using the Haystack annotation tool. Four annotators were employed for the task of creating question-answer pairs.

1. A user visits the address of the AI tool.
2. A user is prompted to log into the AI tool using Open Authentication.
3. A user lands on the welcome page of the AI tool which describes the purpose of the tool.
4. A user makes a query to the AI tool through natural language in the provided message box.
5. The AI tool responds to the user’s query in natural language.

Smallholder farmers and extension officers.

Text data.

An answer in form of text.

An option for the user to see from which passage the answer is from.

No.

It can provide on-farm advice overcoming geographical barriers. The AI tool can be easily expanded to serve more users and integrated with other services. Farmers receive immediate advice allowing for quick decision-making.

Not measured.

The AI tool can provide instant responses to farmer queries. The AI tool is designed to communicate in Swahili language ensuring ease of use. Farmers can access the tool via mobile devices such smartphones.

Yes.

https://mega.nz/folder/p1c0XBiL#C9J5zG6aG7sHAdN6OhtFxg

None.

The AI tool is only accessible through a web-based prototype and requires internet connectivity.

The AI tool responses are limited to single turn interaction.

Potential biases in the AI tool responses due to dataset imbalance.

The AI tool should be hosted on a platform with advanced GPU capabilities.

More smallholder farmers, extension officers and agronomists should be involved in testing and validating.

Not available

The measured accuracy of the AI tool is 73.6%.

Longer response time due hardware capabilities.

The AI tool can make mistakes in responses given.

How many smallholder farmers are using the AI tool?

What is the user satisfaction rate regarding the relevance and accuracy of information provided?

How does the AI tool influence decision-making in horticulture farming?

How do costs of deploying and maintenance the AI tool compare to traditional extension services?

The AI tool may lack understanding of the Swahili language to the extent of which an extension officer can.

The AI tool responses depend on trained data which can become outdated and less applicable.

Data limitations.

Connectivity challenges

Potential model biases due to dataset imbalance.

Track usage metrics and frequency of interaction.

User feedback mechanism to understand user experiences and satisfaction.

In-depth case studies to understand the tool’s contribution to improved horticulture outcomes.

The AI tool can be deployed as a virtual extension officer.

Analyzing data from user interaction to identify trends in horticultural challenges.

Use feedback and performance metrics to evaluate effectiveness of implemented policies.

Support planning for training programs by identifying gaps in farmers’ knowledge and misunderstood practices.

Improved access to horticultural information.

Reduces the need for frequent and expensive visits by extension officers.

Increased productivity and profitability.

The tool can be tailored to other agricultural sub-sectors.

The tool can be customized to be accessible via Short Messaging Service (SMS) to ensure usability in areas with limited internet connectivity.

Improved access to extension services.

Enhanced horticultural productivity and sustainability.

Scaling and expanding.

Implement adaptive learning of the AI tool based on user feedback.

Short Messaging Service (SMS) support.

Integration with services like agricultural market places and price feeds.

Comprehensive user guide and FAQs to help users.

Government support for widespread adoption and scaling.

Collaboration with research institutions to improve the quality of the datasetfor training the AI tool.

Promote the tool’s use as part of the effort to improve horticultural practice.