How AI models in climate science, agriculture, and software systems are shaping the future of development—and what developers should focus on. And finally, the cross-cutting trends that might pique your interest as a developer.
Artificial Intelligence (AI) is increasingly becoming part of the core infrastructure for development. Whether it's being used to monitor the climate, manage agriculture, or power intelligent software systems, AI is proving to be a practical tool—not just a buzzword.
In this post, I’ll summarise three insightful projects led by African developers and researchers. These examples show how AI is already being used to address real-world challenges. More importantly, they offer developers practical lessons on tools, methods, and approaches to building impactful solutions.
1. Smarter Earth Monitoring with Geospatial Foundation Models
Presented by: Wanjiru Catherine, IBM Research Africa
This project introduced Geospatial Foundation Models (GFMs) built using self-supervised learning techniques. These models are designed to understand Earth Observation data such as satellite imagery. They’re trained on diverse geospatial tasks like land cover classification, flood detection, or estimating biomass.
The team also built TerraTorch — a toolkit that makes it easier to fine-tune these models. Built on top of PyTorch Lightning and TorchGeo, TerraTorch allows you to train models for tasks like image segmentation or classification without writing much boilerplate code.
Why this matters for developers:
- You can use pre-trained models to solve different problems by fine-tuning rather than starting from scratch.
- TerraTorch supports a config-only approach to training pipelines, making it friendly for experimentation.
- These models make it easier to work with satellite data and apply AI to environmental and sustainability issues.
Explore it: GitHub – TerraTorch Examples
2. Predicting Maize Pest Outbreaks with Weather Data
Presented by: Edward Gichura, Dedan Kimathi University of Technology
This project focuses on improving food security using machine learning. Edward built models that predict pest and disease outbreaks in maize farms using seasonal weather data such as rainfall, humidity, and temperature.
Using XGBoost, the best model reached 96% accuracy in predicting when and where pests like Fall Armyworm would appear. The datasets combined weather APIs with on-the-ground reports from farmers and agricultural agencies.
Why this matters for developers:
- You don’t always need deep learning. With the right data and feature engineering, classical models like XGBoost can do an excellent job.
- The project makes a strong case for building early warning systems that could be delivered through SMS or mobile dashboards—tools that smallholder farmers can actually use.
- It's a great example of applying AI in a resource-limited setting where impact matters more than hype.
Tip: Consider combining weather APIs with real-time field data for ongoing model retraining.
3. Building Intelligent Agents with Google ADK
Presented by: Ephraim Mwereza
This project was about creating intelligent software agents using Google’s AI Development Kit (ADK). These agents are designed to understand spoken input, manage dialogue, and respond based on context.
The aim here was to build agents that could work even in low-connectivity areas. While the technical details were limited, the emphasis was on creating tools that are both smart and accessible—for example, an agent that can help with education or healthcare tasks.
Why this matters for developers:
- Intelligent agents are more than just bots—they can manage state, handle complex tasks, and learn from interactions.
- Tools like Google ADK allow you to build voice-first applications that are practical in areas where typing isn’t ideal.
- There’s a growing need for lightweight, on-device AI that can run offline or with minimal internet access.
Cross-Cutting Trends That Developers Should Note
Looking across all three projects, a few common themes stand out:
1. Simpler Workflows Through Low-Code Tools
Whether it’s TerraTorch or Google ADK, developers now have access to tools that simplify the training and deployment of AI models. This reduces the barrier to entry and allows for quicker iteration.
2. Pre-Trained Models Are the New Starting Point
Instead of building models from the ground up, developers can start with a general-purpose model and fine-tune it for their own use case. This approach is efficient and scalable.
3. AI Needs to Work in the Real World
These projects serve communities that don’t always have high-speed internet or powerful devices. As a developer, you’ll need to design for edge deployment, offline support, and clear user feedback.
4. Developers Must Be Domain-Aware
You can’t just build in isolation. Understanding the domain—whether it’s agriculture, climate science, or public health—is essential if you want your solution to be useful.
Final Thoughts
These projects are strong examples of how AI can be applied to development challenges in real, practical ways. They show that it’s not just about cutting-edge algorithms, but about building tools that work—tools that people can trust and use.
If you're a developer looking to get involved in socially impactful AI, these case studies offer both inspiration and direction. You already have the technical skills—what matters now is how you choose to apply them.
Want to explore more?
If you found this post useful, consider following for more articles on applied AI, low-code tooling, and building for impact.
Top comments (0)