AgriVoice — Voice AI for Nigerian Farmers at Wetech AI Hackathon

Most agricultural AI solutions are built for farmers who don't exist — farmers with smartphones, reliable internet, and digital literacy. At the Wetech AI Hackathon, Zeroday built AgriVoice — a multilingual voice-AI assistant accessible via basic phone and USSD, designed for the farmers who actually need it.

The gap in agricultural technology isn't about the technology itself — it's about who the technology is built for. Most agri-tech startups design for the 10% of farmers who are digitally connected. The other 90% — smallholder farmers in rural Nigeria — are left behind.

These farmers face real problems: crop diseases, pest outbreaks, changing weather patterns, market price fluctuations. They need agricultural expertise. But they can't access it through apps they can't use, in languages they don't speak, on devices they don't own.

70%
Rural Farmers
500+
Local Languages
48h
Build Time
USSD
Primary Channel

The Problem — Why AgriTech Must Be Inclusive

The Wetech AI Hackathon theme was AI for social impact. When we looked at agriculture, we saw a massive inclusion problem. The most sophisticated agricultural AI systems — computer vision for disease detection, predictive analytics for yield optimization, market intelligence platforms — all assume smartphone access and digital literacy.

But the reality of Nigerian agriculture is different:

Language barriers. Many farmers speak local languages — Hausa, Yoruba, Igbo, and dozens of others. Most agri-tech tools are in English or require literacy to navigate menus and read outputs.

Connectivity gaps. Rural areas often have poor or no internet coverage. Data-heavy apps simply don't work. Even when connectivity exists, data costs are prohibitive for farmers operating on thin margins.

Device limitations. Many farmers use basic feature phones, not smartphones. Touch interfaces, app stores, and complex navigation are foreign to them.

"The best agricultural AI is useless if the people who need it can't access it."

We asked ourselves: what if we built agricultural AI that meets farmers where they are — on their phones, in their languages, using the connectivity they have?

That question led to AgriVoice.

What We Built

AgriVoice is a voice-first agricultural assistant that works through two primary channels: regular phone calls and USSD (the *#* codes that work on any phone, even without data).

The Build — Voice AI & USSD Integration

Our backend was Django with PostgreSQL. The core challenge was the voice AI integration. We used a combination of speech-to-text, natural language processing, and text-to-speech services, all optimized for low-bandwidth environments.

For language support, we trained custom models on agricultural terminology in Hausa, Yoruba, and Igbo. This wasn't just translation — it was understanding agricultural concepts in local dialects and contexts. "Yellow leaves" might mean nitrogen deficiency in one context and water stress in another — the AI had to understand the difference.

The USSD integration was technically simpler but UX-critical. We designed the menu hierarchy carefully — no more than three levels deep, with clear numeric options and the ability to go back at any point. We tested the flow with actual farmers to ensure it felt natural, not like a technical system.

"USSD isn't a fallback — for many farmers, it's the primary interface. Design it like it matters."

What We Learned

Lesson 01
Voice AI must understand context, not just words.

Early versions of our voice model would translate "my tomatoes are dying" literally and give generic advice. But farmers don't speak like textbooks — they describe symptoms in local terms, with cultural context. We had to train the AI to understand agricultural language as it's actually spoken, not as it's written in manuals.

Lesson 02
Low-literacy design is different from simple design.

Simple design assumes the user can read. Low-literacy design assumes they might not be able to. Voice-first interfaces, audio feedback, and visual cues (like colored buttons on feature phones) become essential. We learned to test with users who had varying literacy levels, not just the digitally savvy.

Lesson 03
Offline capability isn't optional in rural Nigeria.

We built caching mechanisms so that frequently accessed information — like common pest treatments or standard planting calendars — could be retrieved without an active data connection. The system syncs when connectivity is available but remains functional when it's not.

Lesson 04
Trust is built through human connection, not just accuracy.

Farmers were initially skeptical of AI recommendations. What convinced them wasn't the technology — it was the ability to connect with a human extension officer when the AI wasn't enough. The AI handles routine queries; humans handle the complex ones. That hybrid approach built trust.

What's Next for AgriVoice

AgriVoice is currently in pilot testing with farming cooperatives in three states. Early feedback has been positive — farmers appreciate being able to get answers in their own language without traveling to extension offices or waiting for agricultural officers to visit their communities.

We're expanding the language support to include more local dialects and adding more specialized agricultural domains — livestock management, aquaculture, and post-harvest storage. We're also exploring partnerships with agricultural input suppliers to connect farmers with seeds, fertilizers, and equipment through the same voice interface.

"Agricultural technology should amplify farmer wisdom, not replace it. The best AI learns from farmers, not just teaches them."

If you're interested in deploying AgriVoice in your farming community or contributing agricultural expertise to our knowledge base, get in touch. We're building for the millions of farmers who feed Nigeria — and they deserve technology that speaks their language.

Want to support AgriVoice?

We're looking for agricultural experts, language specialists, and pilot partners.

Follow @PrinceJetro123 View All Projects