Cloud Computing Projects in Bangalore

An event-driven image processing pipeline is built on AWS using S3, Lambda, and DynamoDB. When a user uploads an image to an S3 input bucket, a Lambda function is triggered to resize the image to three standard dimensions, apply watermarking, and store the processed versions in an output bucket. Image metadata and processing status are recorded in DynamoDB. An API Gateway REST endpoint allows clients to query processing status by image ID. The pipeline handles 1,000 concurrent uploads without configuration changes, demonstrating serverless auto-scaling. Infrastructure is provisioned with AWS SAM templates.

A three-tier web application — React frontend, Node.js API layer, and PostgreSQL database — is deployed on AWS with production-grade architecture. The frontend is served from S3 via CloudFront CDN. The API layer runs on EC2 instances behind an Application Load Balancer with an Auto Scaling Group configured to scale between 2 and 10 instances based on CPU utilisation. The database runs on RDS PostgreSQL with a Multi-AZ standby for high availability. A load test using Apache JMeter demonstrates the auto-scaling behaviour under 500 concurrent users. Terraform provisions the entire infrastructure.

A real-time analytics pipeline ingests simulated IoT sensor events at 10,000 events per second using Apache Kafka (local) and AWS Kinesis Data Streams (cloud). A Kinesis Data Analytics application running Apache Flink computes 1-minute windowed aggregates — average temperature, maximum vibration, and anomaly count — and writes results to DynamoDB. A Grafana dashboard connected to DynamoDB via a Lambda data source displays live metrics with sub-second latency. The project benchmarks Kafka and Kinesis on throughput, latency, and cost per million events.

A CI/CD pipeline is built for a three-service microservices application (user service, order service, notification service) using GitHub Actions, Docker, Amazon ECR, and Amazon EKS. On every push to main, GitHub Actions runs unit tests, builds Docker images, pushes them to ECR, and deploys to an EKS cluster using Helm charts. Blue-green deployment is configured to achieve zero-downtime releases. The pipeline includes a security scanning stage using Trivy to detect container image vulnerabilities. End-to-end pipeline execution time from push to production is benchmarked at 4.2 minutes.

A machine learning workflow is implemented on Google Cloud Platform using Vertex AI. A custom training job trains a ResNet-50 image classifier on the CIFAR-10 dataset using a GPU-accelerated Vertex AI Training instance. The trained model is registered in Vertex AI Model Registry and deployed to a Vertex AI Endpoint for online prediction. A Cloud Run service provides a REST API wrapper around the endpoint. Cloud Monitoring dashboards track prediction latency, error rate, and model accuracy over time. The project includes a cost analysis comparing Vertex AI to a self-managed GKE deployment.

A pilot-light disaster recovery architecture is designed and implemented for a web application across AWS us-east-1 (primary) and ap-south-1 (Mumbai) regions. The primary region runs a full application stack; the DR region maintains a minimal warm standby with RDS read replica and a stopped EC2 Auto Scaling Group. Route 53 health checks monitor the primary region and automatically fail over DNS to the DR region within 60 seconds of a primary outage. A simulated failover test demonstrates RTO of 3.5 minutes and RPO of 45 seconds. The architecture is documented with a runbook and cost breakdown.

A food delivery application is decomposed into five microservices — authentication, restaurant catalogue, order management, payment processing, and notification — each containerised with Docker. The services communicate via REST APIs and an NATS message broker for asynchronous events. The application is deployed on a local Kubernetes cluster (minikube) with Helm charts, horizontal pod autoscaling, and liveness/readiness probes. An Ingress controller routes external traffic to the appropriate service. The project includes a chaos engineering exercise using Chaos Monkey to validate service resilience under pod failures.

A cost visibility and optimisation tool ingests AWS Cost and Usage Reports from S3, processes them with a Python ETL pipeline running on Lambda, and stores aggregated cost data in DynamoDB. A React dashboard displays cost breakdowns by service, region, and tag with month-over-month trend charts. An anomaly detection module using Z-score analysis flags unusual cost spikes and sends SNS email alerts. The tool identifies three optimisation opportunities — right-sizing over-provisioned EC2 instances, deleting unattached EBS volumes, and switching to Reserved Instances — with estimated annual savings of $4,200 for a simulated workload.