When building scalable platforms for agriculture, managing hundreds of APIs and concurrent user states can quickly become a bottleneck. In our project AgriProject (often referred to as FarmFlow), we decided to adopt a microservices architecture right from the start to ensure high availability, clean separation of concerns, and seamless integration of Machine Learning models.
Why Spring Boot for the Backend?
Spring Boot remains the gold standard for enterprise backend development. Its massive ecosystem and built-in support for Cloud-native patterns made it an obvious choice to handle our 120+ REST APIs. We utilized:
- Spring Cloud Eureka for Service Discovery, allowing our backend components to dynamically find each other.
- Spring Security + JWT for stateless, secure authentication across all services, with role-based access for Admins, Farmers, and Users.
- Caffeine Cache for high-performance, local caching of static geographical and crop data.
- HikariCP for robust database connection pooling against our MySQL instance.
Integrating Python ML Models
One of the core requirements of AgriProject was running predictive models for smart farming. Instead of trying to force Python ML scripts into a Java environment, we decoupled them:
- Python Flask/FastAPI microservices wrap the Scikit-learn models. We successfully deployed three core models achieving 95%+ accuracy:
- Crop Recommendation (Model 5)
- Smart Pump Control (Model 6)
- Fertilizer Recommendation (Model 7)
- The Spring Boot backend acts as an API Gateway and orchestrator, sending asynchronous requests to the Python services using REST clients.
- Everything is containerized using Docker and orchestrated via Docker Compose for local development, allowing us to spin up MySQL, Eureka (port 8761), the Java Backend (port 8080), and the ML Service (port 5000) with a single command.
// Example of a simple REST call from Spring Boot to the Python ML Service
@Service
public class YieldPredictionService {
private final RestTemplate restTemplate;
public PredictionResponse getPrediction(PredictionRequest request) {
String mlServiceUrl = "http://localhost:5000/predict";
return restTemplate.postForObject(mlServiceUrl, request, PredictionResponse.class);
}
}
The Modern Frontend Experience
To bring this powerful backend to the users, we built a React application using Vite. It features a premium, glassmorphism UI that includes: - A complete e-commerce shopping flow (Cart → Checkout → Orders). - Complex address management supporting all 37 Indian states and UTs. - Interactive dashboards tailored for Users, Farmers, and Admins.
By keeping the frontend completely decoupled from the backend and communicating purely via REST APIs, we achieved a highly scalable and maintainable architecture. Next time, I'll dive deeper into how we transition these containerized services to a Kubernetes cluster with Horizontal Pod Autoscaling!