CloudData Science

CloudEco

Environmental ML Cloud Application

A scalable cloud-native machine learning inference system for environmental object detection using YOLO, FastAPI, Docker, and Kubernetes.

PythonFastAPIYOLODockerKubernetesLocustTerraformREST APILinux

Problem

Environmental monitoring teams needed a scalable, reproducible way to run object detection on large image batches without managing GPU infrastructure manually.

My role

Cloud / ML Engineer — designed the inference API, container strategy, Kubernetes manifests, and load-testing harness.

Solution

Built a FastAPI service wrapping YOLO, containerised with Docker, deployed to Kubernetes with horizontal pod autoscaling, and provisioned via Terraform. Locust tests benchmarked p95 latency at 1/2/4/8 pods.

Challenges

›Balancing inference latency against pod cost at multiple replica counts
›Reliable Base64 image transport over REST
›Reproducible infrastructure across environments

Key features

›RESTful API for image inference
›Base64 image processing
›YOLO-based object detection
›Docker containerisation
›Kubernetes deployment with autoscaling
›Load testing with Locust
›Performance benchmarking across 1, 2, 4, and 8 pods
›Infrastructure-as-Code using Terraform

Results

›Linear throughput scaling up to 8 pods
›Repeatable Terraform-driven environments
›Documented performance envelope for capacity planning

Cloud-native deployment

Horizontal scalability

Production-grade API engineering