Architecture Diagram of the AI-Powered FarmDepot Agric Expert Agent
Mobile View of the FarmDepot Marketplace (website)
Mobile View of the AI Agent (Mama FarmDepot) integrated into the FarmDepot website.
Desktop View of the Farmdepot Agric Expert Agent as a closed widget at the bottom right corner
Desktop View of the Farmdepot Agric Expert Agent as opened widget ready for conversation
Desktop View of the Farmdepot Agric Expert Agent as a closed widget at the bottom right corner integrated in farmdepot.ng for live use
Desktop View of the Farmdepot Agric Expert Agent as opened widget integrated into farmdepot.ng for live use
Farmdepot mockup

Farmdepot Agric Expert Agent

Inspiration

In many rural communities across Nigeria, farmers and traders remain excluded from digital marketplaces, not because they lack products, but because they lack literacy, language access, and digital skills.

While working closely with farmers through FarmDepot, I witnessed a recurring problem:

Farmers had goods to sell but couldn’t navigate apps, type search queries, or communicate in English.

This created a paradox:

Food Supply Exists ≠ Market Access

The inspiration behind this project was simple but powerful:

What if farmers could just speak in their native language and access the market instantly?

What it does

The FarmDepot Agric Expert Agent is an AI-powered system that enables farmers and traders to:

Post agricultural products using voice

Search for products via speech or text

Interact in multiple Nigerian languages (English, Hausa, Yoruba, Igbo)

Receive voice feedback (Text-to-Speech)

Access a simple, user-friendly dashboard

Send email notifications to the company's email for follow-up and other necessary human actions

It transforms the marketplace into a conversation, removing barriers of literacy and language.

How we built it

We designed the system as a multimodal AI agent combining speech, language, and data systems:

Core Technologies

Frontend/UI: WordPress & React

Backend: Node.js and Express

Database: Firestore

Speech Recognition: Gemini APIs

Voice Output: Google Gemini engine

Multilingual Layer: Language detection + translation pipelines (Gemini)

Hosted on Google Cloud Run

Email notifications: Nodemailer/SMTP

CI/CD and Version Control: GitHub

Key Features Implemented

Voice-based product posting

Voice-enabled search

Multilingual interaction pipeline

Structured product database (users & listings)

Lightweight dashboard for interaction

Challenges we ran into

1. Speech Recognition Accuracy

Local accents and dialects significantly affected transcription accuracy.

2. Multilingual Complexity

Handling multiple languages required:

Language detection

Translation consistency

Context preservation across languages

3. Connectivity Constraints

Rural users often operate in low-bandwidth environments, affecting real-time voice processing.

4. Integration Issues

Combining voice, translation, and database operations into a seamless pipeline required careful orchestration.

Accomplishments that we're proud of

✅ Built a fully functional voice-enabled marketplace

✅ Enabled non-literate users to interact digitally

✅ Successfully integrated multilingual AI interaction

✅ Created a real-time working prototype (no mockups)

✅ Impacted early users with improved access to buyers

What we learned

This project reinforced key lessons: AI is most powerful when it is inclusive

Simplicity beats complexity in real-world adoption

Local context (language, culture) is critical for AI success

Building for underserved users requires empathy, not just technology

We also deepened our expertise in: Speech systems integration

Multilingual AI pipelines

User-centered design for low-tech environments

What's next for FarmDepot Agric Expert Agent

Phase 3 Roadmap

Add more local languages (Ebira, Igala, fufulde, Ibibio, Gwari, Tiv, Idoma, and Larntan) Improve speech models with localized datasets Introduce offline/low-bandwidth voice capabilities