Cloud-Native Deployment for Real-Time Agritech Disease Diagnostics

Problem Statement

Farmers uploading high-resolution leaf images encountered slow upload speeds and timeout errors due to remote server locations, leading to frustration and low tool adoption.

Key Challenges

• !Large upload file sizes from modern phone cameras choking bandwidth.
• !High inference server costs when processing image analysis requests.
• !Unpredictable rural network connectivity.

The Solution

We deployed Edge Serverless Workers to handle edge requests and compress uploaded images at the edge. The disease classification model was converted to TensorFlow Lite format to run client-side whenever possible, falling back to a containerized GPU cluster on AWS ECS.

Our Development Process

1. 1

   Phase 1

   Custom image compression algorithms implementation at the client side.

2. 2

   Phase 2

   GPU-enabled container configuration for TensorFlow model hosting.

3. 3

   Phase 3

   Setting up geo-distributed edge endpoints via global CDN.

4. 4

   Phase 4

   Integrating offline diagnostic sync using local SQLite caching.