MavunoSasa - Smart harvest decisions

MavunoSasa is a predictive platform that empowers farmers with early warnings on food price spikes, enabling smarter decisions, timely interventions, and food security in Africa.

Comment

🌾 MavunoSasa – Smart Harvest Decisions

Early warnings for food price spikes, built for Africa’s most vulnerable.

Inspiration

In 2020, I stepped away from my tech work and ventured into farming in the Mau Forest, driven by the promise of high returns on potatoes. I spent four months immersed in the wild, pouring in time, money, and heart. But on harvest day, everything unraveled! Market prices had crashed, fertilizer costs had spiked unpredictably, and the weather was erratic. I was devastated.

My co-founder Brian has seen these realities time and again. Families forced to choose between school fees and supper due to price shocks. Millions in Kenya live this story every harvest season.

MavunoSasa was born from this pain: a deep-tech platform to give smallholder farmers and aid agencies a fighting chance by predicting price shocks before they hit.

What it does

MavunoSasa is a food price early warning system that predicts commodity price spikes 2–3 months in advance using market data, climate indicators, and economic trends.

It empowers:

  • Farmers to make smart planting and selling decisions
  • Aid organizations to time procurement more effectively
  • Governments to activate food aid or subsidies earlier

Users access insights through:

  1. A Progressive Web App (PWA) that works offline
  2. SMS alerts for low-data users
  3. USSD menus for basic feature phones (coming soon)

This multi-channel strategy ensures reach across even the most digitally disconnected rural communities.

🛠️ How we built it

Tech Stack

  • Backend: Python, FastAPI, PostgreSQL with PostGIS, Docker
  • ML Models: XGBoost, ARIMA, Prophet, LSTM
  • Frontend: React PWA with offline access (Service Workers), Streamlit dashboards
  • Data Sources:
    1. WFP food prices
    2. CHIRPS rainfall data
    3. KNBS economic data
    4. OpenStreetMap geodata

Modeling

We trained classification and regression models to:

  • Predict major price spikes:

Predicting Function

  • Forecast price changes
  • Generate crop recommendations based on market and climate signals

Challenges we ran into

Messy market data: duplicate names, missing units, inconsistent formats

Cross-source merging: aligning monthly price data with daily weather readings

Sparse rural coverage: limited historical info for some remote regions

Designing for extremes: building tools usable by both low-tech farmers and policy analysts

Lightweight intelligence: ensuring models perform well even offline or with poor connectivity

USSD code acquisition: getting a USSD short code on time proved challenging due to telco onboarding delays and regulatory bottlenecks

Accomplishments that we're proud of

✅ Built a functional MVP that works offline and supports constrained rural environments

✅ Developed a dual-purpose interface usable by both policy experts and smallholder farmers

✅ Achieved early warning forecasts that outperform naive trend models by over 25% accuracy

✅ Consolidated and cleaned over 15 years of food price data from fragmented market sources

✅ Integrated climate data and geospatial mapping into our prediction pipeline

✅ Prototyped SMS alerts and laid groundwork for USSD integration, despite delays in short code acquisition

What we learned

  • The most powerful AI is the most accessible
  • Localizing tech for Africa means solving real-world constraints, not just building features
  • Timing is everything for both farmers and aid organizations
  • Simplicity, transparency, and offline-first design are essential

What's next for MavunoSasa - Smart harvest decisions

  • Launch localized SMS alerts: We’re refining our notification system to deliver timely alerts to farmers via SMS including crop-specific price forecasts and smart harvest recommendations. Pilots will start in potato-growing zones in Kenya.

  • Roll out USSD support (despite initial delays): Acquiring a USSD short code has proven time-consuming due to telco onboarding and licensing requirements. However, the USSD interface is already prototyped and will be activated as soon as telco access is secured ensuring accessibility even for basic feature phones.

  • Partner with food policy stakeholders: We are hoping to initiate conversations with key players like the Ministry of Agriculture, FAO, and WFP Kenya to align MavunoSasa’s insights with national food policy planning, subsidy deployment, and early-response frameworks.

  • Open-source the prediction engine: To accelerate collaboration, transparency, and external validation, we plan to open-source our core price prediction and crop recommendation modules. This will support researchers, NGOs, and agritech developers across Africa.

  • Scale regionally in East Africa: With Kenya as our launchpad, we plan to expand to Uganda, Tanzania, and Rwanda, adapting MavunoSasa to new market structures, crop types, and climate zones building toward a regional food price intelligence system that thrives under constraint.

MavunoSasa isn’t just software.
It is a system to transform food security by making our continent not just prepared, but predictive.

Built With

Share this project:

Updates