SokoLeo

Inspiration

Smallholder farmers across Kenya often make farming decisions without access to reliable market data, weather insights, or logistics guidance. This leads to low profits, oversupply in markets, and significant post-harvest losses.

We were inspired to build SokoLeo to bridge this gap by giving farmers access to real-time, data-driven insights through a simple conversational interface. Instead of guessing, farmers can now make informed decisions on what to plant, when to harvest, and where to sell.

What it does

SokoLeo is an AI-powered agricultural assistant that helps farmers:

Identify the most profitable crops to plant
Access real-time market prices and demand
Get recommendations on where to sell produce
Optimize transport routes and logistics
Plan based on weather patterns
Reduce losses through better storage practices

It acts like a smart farming advisor, combining multiple sources of intelligence into one simple system.

How we built it

We built SokoLeo using a multi-agent AI architecture, where each agent handles a specific task:

Market Agent → Fetches real-time prices and demand
Weather Agent → Provides planting and harvesting guidance
Logistics Agent → Suggests transport and storage strategies
Profit Agent → Evaluates profitability of crops
Recommendation Agent → Combines all insights into one answer

These agents are orchestrated using LangGraph, allowing dynamic routing of farmer queries.

Tech Stack

Python
LangGraph (multi-agent orchestration)
Tavily API (real-time search)
Persistent memory (JSON-based storage)
Environment variables for secure API handling

Challenges we ran into

Coordinating multiple agents to produce one clear, unified response
Integrating real-time market data without slowing down the system
Designing context-aware memory so the system remembers previous farmer queries
Keeping the interface simple and usable despite complex backend logic

Accomplishments that we're proud of

Successfully built a fully functional multi-agent AI system
Implemented dynamic routing based on farmer questions
Created a system that provides context-aware recommendations using memory
Delivered a solution that directly addresses a real-world problem affecting farmers

What we learned

Multi-agent systems are powerful for solving complex, real-world problems
Combining different data sources (market, weather, logistics) leads to better decisions
Simplicity in user experience is critical, especially for non-technical users
Building AI for impact requires balancing accuracy, usability, and speed

What's next for SokoLeo

Integrate live market APIs and satellite weather data for higher accuracy
Add SMS/USSD support so farmers without smartphones can access the system
Support local languages to improve accessibility
Build a mobile and web app interface for wider adoption
Partner with markets and buyers to enable direct farmer-to-buyer connections