Back to Projects

Cloud AtlasClassifier

Project Motivation & Problem Statement

Deploying machine learning models in production often introduces significant infrastructure complexity-provisioning servers, managing scaling, and handling operational overhead. Many teams struggle to bridge the gap between a trained ML model and a production-ready API. Cloud AtlasClassifier addresses this by leveraging AWS serverless architecture to create a fully deployable, auto-scaling ML categorization pipeline that eliminates server management while maintaining low-latency inference.

The goal was to build an end-to-end serverless application that could accept input data, run it through a trained classification model, and return categorized results-all without managing any underlying compute infrastructure.

Technical Approach

1. Serverless Architecture Design

  • Designed the entire application around AWS Lambda functions, enabling automatic scaling based on incoming request volume with zero idle costs.
  • Used AWS SAM (Serverless Application Model) CLI for local development, testing, and deployment, providing Infrastructure-as-Code (IaC) through a template.yaml definition.
  • Defined API Gateway endpoints to expose the Lambda functions as RESTful APIs for external consumption.

2. ML Model Integration

  • Integrated a pre-trained classification model into the Lambda function runtime, packaging model weights and inference code together for deployment.
  • Implemented input validation and preprocessing within the Lambda handler to ensure robust handling of varied input formats.
  • Optimized cold-start latency by minimizing package size and lazy-loading heavy dependencies.

3. Application Structure & Testing

  • src/: Core application logic including the Lambda handler, model loading, and prediction functions.
  • events/: Sample invocation event payloads for local testing and debugging with SAM CLI.
  • tests/: Unit tests validating model inference, input parsing, and API response formatting.
  • template.yaml: AWS CloudFormation template defining Lambda functions, API Gateway, IAM roles, and resource permissions.

4. CI/CD and Deployment Pipeline

  • Configured automated build and deployment workflows using SAM CLI, enabling one-command deployment to AWS.
  • Structured the project for continuous integration with local test execution before cloud deployment.
  • Implemented environment-based configuration to support development, staging, and production deployments.

Results

  • Successfully deployed a fully serverless ML categorization pipeline on AWS that auto-scales with demand.
  • Achieved sub-second inference latency for warm Lambda invocations.
  • Eliminated operational overhead of server provisioning and maintenance.
  • Structured codebase enabled rapid iteration and local testing before cloud deployment.

Limitations

  • Lambda cold starts can introduce latency spikes for infrequent requests.
  • Model size is constrained by Lambda's deployment package limits (250 MB unzipped).
  • Serverless architecture may not be cost-effective for extremely high-throughput, sustained workloads.

Skills and Technologies Demonstrated

  • AWS Lambda and serverless architecture design
  • AWS SAM CLI for Infrastructure-as-Code
  • ML model deployment and inference optimization
  • Python, JavaScript, and PowerShell scripting
  • RESTful API design with API Gateway
  • Unit testing and CI/CD pipeline configuration

Resources