Inspiration
Smallholder farmers in Rwanda and across Sub-Saharan Africa lose 20–40% of their harvest to crop diseases annually. These losses translate to approximately $220 billion globally, with the heaviest impact on food-insecure regions.
Key barriers preventing timely disease intervention:
1. Limited Access to Extension Services: Rwanda has approximately 1 extension officer per 1,500 farmers (MINAGRI, 2020), making timely expert consultation nearly impossible for remote farmers.
2. Language & Literacy Barriers: Most agricultural resources are in English or French, while Kinyarwanda is the primary language for 99% of Rwandan farmers. This language gap delays diagnosis and leads to misapplied treatments.
3. Connectivity Gaps: Only 37% of rural Rwandans have reliable internet access (Rwanda Utilities Regulatory Authority, 2022), rendering existing online diagnostic tools unusable for the majority of farmers.
4. Knowledge Gap: Farmers often cannot identify diseases early or confuse symptoms, leading to incorrect treatments, wasted resources, and crop failure.
What it does
VunaLink is a mobile-first, offline-capable crop disease detection app designed specifically for Rwandan smallholder farmers. It combines computer vision AI with localized language support to deliver instant, actionable disease diagnosis—entirely in Kinyarwanda and without requiring internet connectivity.
How It Works
1. Snap a Photo
Farmers photograph an affected leaf or crop using their smartphone camera.
2. Instant AI Diagnosis (Offline)
A lightweight TensorFlow model runs locally on the device, identifying the disease within seconds—even in remote areas with zero connectivity.
3. Results in Kinyarwanda
The app displays:
Disease name
Confidence Score
Text-based AI guidance providing step-by-step treatment instructions and preventive measures for future crops
How we built it
For the technology, we used:
TensorFlow to train the AI model and ONNX Runtime to run the model offline on smartphones and other devices.
MobileNetV2, trained using a selected portion of the PlantVillage dataset, to identify diseases in 6 categories: Potato (Healthy, Early Blight, Late Blight) and Tomato (Healthy, Early Blight, Late Blight).
Progressive Web App (PWA) so farmers can access VunaLink through a browser without needing to download it from an app store.
We designed the interface and guidance with local farmers in mind, using simple language and a user-friendly interface.
The app has three core features: disease detection (take or upload a photo and get an instant diagnosis), treatment guidance (receive practical step-by-step recommendations), and activity logging (keep a record of previous diagnoses and build a history for each farmer).
Finally, we built a working prototype that demonstrates the complete user flow, from capturing or uploading a crop image to disease detection and receiving treatment recommendations.
Challenges we ran into
1. Building an AI Model That Works Offline: Our biggest technical challenge was creating a disease detection model that could run entirely offline on low-end smartphones. We had to evaluate various AI models before finding one that was lightweight and could run offline effectively.
2. Designing for Kinyarwanda Users: Research revealed that most smallholder farmers in Rwanda do not speak English, so all disease names, treatment instructions, and AI guidance had to be delivered in clear, simple Kinyarwanda. The challenge was ensuring the language was practical and actionable for farmers managing their crops in real time.
3. Limited Access to Local Training Data: While the PlantVillage dataset provided a strong foundation, the large volume of data was slowing the system down and significantly increasing the app size. We selected a focused subset—6 disease categories across potato and tomato—to keep VunaLink lightweight and fast for offline use.
4. Balancing Features vs. Simplicity: Early on, we wanted to add weather forecasts, market prices, and community forums. The hardest challenge was saying no to good ideas and staying focused on the core problem: instant, offline disease detection. Cutting features felt like losing value, but it was essential to building something farmers would actually use.
Accomplishments that we're proud of
1. We Built a Fully Offline AI Solution
VunaLink works without any internet connection. By optimizing a MobileNetV2 model and exporting it to ONNX for offline inference, we created a lightweight AI that runs locally on the device, delivering results in seconds. This makes the app accessible to farmers in remote areas where connectivity is unreliable.
2. We Designed for the User, Not the Technology
Every design decision — from simple interface to Kinyarwanda language — was made with Rwanda's smallholder farmers in mind. We started with the farmer's reality and built the technology around it, not the other way around.
3. We Validated a Real Problem with Measurable Impact
Through research, we confirmed that 30–40% of smallholder harvests are lost to preventable diseases. VunaLink has the potential to reduce crop loss by 15–25% and save farmers 2–3 hours per disease incident by eliminating travel to extension officers.
4. We Created a Scalable Business Model
VunaLink isn't just a social impact project; it's a viable business. Our three-phase roadmap shows how we'll grow from a free diagnostic tool into a revenue-generating platform while keeping farmers as free users and monetizing through B2B partnerships.
5. We Built a Foundation for Financial Inclusion
By logging every diagnosis and treatment, we're creating verifiable farmer profiles that banks can use to offer loans without requiring land titles. This could unlock credit access for millions of smallholder farmers across East Africa.
6. We Stayed Focused and Shipped:
In a short timeframe, we went from research to a working prototype with core features: disease detection, treatment guidance, and activity logging. We prioritized and delivered a functional, user-centered solution.
What we learned
Throughout the development of VunaLink, we gained critical insights across research, ideation, and technical execution that shaped our approach to solving a complex agricultural challenge.
During Research:
We learned that 30–40% of smallholder harvests in Rwanda are lost to preventable crop diseases, yet the agricultural extension system is severely overstretched with only 1 officer per 1,000+ farmers. Farmers often travel hours for diagnosis, and by the time they receive help, the damage is irreversible. We also discovered that most existing agritech solutions fail in rural contexts because they require consistent internet connectivity, are built in foreign languages, or assume high digital literacy. This research validated that the real barrier isn't access to technology, it's access to offline, localized, and actionable solutions.
During Ideation:
We learned that solving one problem well is more impactful than building a feature-heavy platform. Early on, we considered adding weather forecasting, market prices, and community forums, but our research revealed that instant disease diagnosis was the most urgent unmet need. We also learned that language and literacy are non-negotiable design constraints. Kinyarwanda support is critical. This insight led us to prioritize text-based AI guidance, ensuring accessibility while keeping the app lightweight.
During Building:
We learned that offline AI is both a technical challenge and a competitive advantage. Training a TensorFlow model that runs on basic smartphones required aggressive optimization, balancing accuracy with file size and inference speed. We also learned the importance of choosing the right training data. We used the PlantVillage dataset with a MobileNetV2 model architecture, which includes thousands of labeled images of crop diseases, allowing us to build an accurate model without needing to collect our own training data from scratch. Building it as a Progressive Web App (PWA) meant farmers could access VunaLink through a browser without app store downloads, removing another barrier to adoption.
Most importantly, we learned that farmers don't need "smart agriculture", they need tools that work in their reality. VunaLink is designed for intermittent connectivity, basic smartphones, and users who've been excluded from the digital agriculture. This user-first philosophy will guide every feature we build moving forward.
What's next for VunaLink
Phase 1: MVP — Rwanda Launch (Months 1–12)
Our immediate priority is launching VunaLink in Rwanda with full Kinyarwanda support and three core features:
Disease Detection — Offline AI identification covering 6 disease categories: Potato (Healthy, Early Blight, Late Blight) and Tomato (Healthy, Early Blight, Late Blight).
Treatment Guidance — Step-by-step, text-based recommendations prioritizing low-cost organic treatments and prevention care.
Activity Logging — Every diagnosis is recorded, building a verifiable history of each farmer's crop health and management practices
Goal: Build deep farmer engagement and capture high-quality data that proves the platform's value. Success metrics include 1,000 active users, 5,000+ diagnoses, and measurable reductions in crop loss (target: 15–25% improvement in early disease intervention).
Phase 2: Regional Expansion + Revenue Activation (Months 13–24)
Once we've validated product-market fit in Rwanda, we'll expand to Kenya (Swahili), Uganda (Luganda), and Tanzania, adapting the AI model and interface for each region's dominant crops and languages.
At this stage, we activate two revenue streams:
Marketplace Integration — Connect farmers directly to buyers (cooperatives, wholesalers, food processors), taking a small transaction fee or subscription from buyers who gain access to verified farmer networks.
Cold Storage Brokerage — Partner with cold storage providers to offer farmers affordable post-harvest storage. VunaLink earns a commission on bookings, helping farmers reduce spoilage while unlocking a new revenue channel.
Goal: Reach 5,000+ farmers across East Africa and achieve break-even on operational costs through marketplace and storage commissions.
Phase 3: Credit Access Enablement (Months 25–36)
The final phase unlocks financial inclusion — the highest-impact, most scalable opportunity in our roadmap. By this stage, VunaLink will have aggregated rich farmer profiles that include:
Crop health history (diseases identified, treatments applied, outcomes)
Marketplace activity (crops sold, buyers engaged, transaction volume)
Cold storage usage (harvest preservation, quality management)
We'll partner with banks, microfinance institutions (MFIs), and fintech lenders to offer loans based on platform activity rather than land titles — a critical innovation, since most smallholder farmers lack formal collateral. VunaLink earns referral fees (1–3% of loan value) when farmers successfully access credit.
Goal: Enable 1,000+ farmers to access their first formal loans, unlocking capital for inputs, equipment, and farm expansion. This phase transforms VunaLink from a diagnostic tool into a financial infrastructure layer for smallholder agriculture.
