AI-Driven Vehicle Classification on AWS for a Defense Agency

D3Clarity partnered with a European defense agency to deliver rapid AI prototyping using AWS SageMaker, enabling automated vehicle image classification from a centralized Amazon S3 data lake. Leveraging a pre-trained PyTorch ResNet model, the team implemented a secure, cost-governed computer vision solution designed for scalability without over-engineering infrastructure. The engagement established best practices for SageMaker governance, IAM-based data access, and responsible compute management while empowering internal teams through knowledge transfer and documentation. The result was a validated cloud-native AI foundation on AWS, providing both immediate operational value and a clear path toward production-grade MLOps and advanced computer vision capabilities.

CASE STUDY

4-MINUTE READ

AI-powered vehicle classification model running on AWS SageMaker

The Challenge at Hand

A European defense agency sought to accelerate its use of AI-powered computer vision to automatically classify vehicle types (e.g., cars, trucks, bicycles) from images stored in a centralized data lake. With a labeled dataset of ~5,000 vehicle images, the agency aimed to leverage AWS-native machine learning services to build an automated classification capability.

However, the engagement faced several constraints:

Limited in-house ML engineering capacity and expertise, with a requirement for rapid time-to-value through a functional pilot.
AWS environment configuration needs, including secure IAM-based access to Amazon S3 and responsible compute resource management to avoid unexpected cloud spend.
The solution had to deliver immediate business outcomes while laying a scalable foundation for future enhancements — such as object detection, custom model tuning, and production-grade MLOps — without prematurely over-engineering infrastructure.

The agency also needed knowledge transfer and enablement to allow internal teams to continue experimenting and iterating independently. This demanded hands-on guidance for configuring Amazon SageMaker services and establishing governance and cost visibility within their AWS environment.

Innovative Solutions Unleashed

D3Clarity designed and implemented a lightweight, cost-conscious AWS ML solution centered on Amazon SageMaker for model development and inference, and Amazon S3 for secure image storage. To accelerate delivery and accommodate limited data volume, D3Clarity chose a pre-trained PyTorch ResNet image classification model, enabling rapid deployment of a functional model while preserving extensibility for future enhancements.

Key steps included:

Focused requirements workshop to validate objectives, success criteria, AWS usage constraints, and dataset readiness — ensuring alignment with organizational goals.
Governed SageMaker Studio setup inside the client’s AWS account, with:
- Resource-level controls
- Tagging standards for cost transparency
- Scoped access policies to avoid unintended compute usage
Secure IAM-based integration between SageMaker and S3 for controlled image access.
A streamlined pilot workflow:
- Ingestion of vehicle images from S3
- Execution of the ResNet classification model in SageMaker
- Generation of structured prediction output (TXT/CSV) back to S3
- Controlled startup/shutdown of SageMaker compute to minimize idle charges (a validated cost-aware pattern)

To empower the client’s internal teams, D3Clarity provided post-handover enablement, including documentation, recorded demos, and consultative support. Recommendations included SageMaker configuration best practices, cost optimization strategies, and architecture guidance for future expansion toward production-grade MLOps and advanced vision services.

AI-powered vehicle classification model running on AWS SageMaker

The architecture was intentionally modular, providing a path to integrate more sophisticated AWS computer vision services, automated pipelines, and object-detection capabilities.

Transformative Results Achieved

D3Clarity delivered a fully functional automated vehicle classification solution on AWS within a short engagement window, demonstrating technical feasibility while aligning with the agency’s cost and schedule constraints.

Key outcomes include:

Validated reference architecture for cloud-native ML on AWS, incorporating governance, secure IAM access, and cost-aware resource management.
A working pilot that proved automated image classification is achievable with existing data and standard AWS tools — reducing technical uncertainty and enabling informed next steps.
Internal team enablement, supported by comprehensive documentation and demo walkthroughs, enabling capability growth through model tuning, alternative algorithms, and evaluation of further AWS services (e.g., SageMaker JumpStart, Autopilot).
A scalable, extensible AI foundation ready to evolve toward production deployment and broader computer vision initiatives, positioning the agency to confidently expand its use of AWS ML services and adopt best practices for secure, governed ML lifecycle management.

This engagement demonstrated how organizations with limited ML resources can rapidly adopt AWS ML services in a governed, cost-aware manner. By leveraging best practices in SageMaker setup, resource governance, and modular architecture design, D3Clarity helped the agency gain both immediate value and a scalable foundation for future AI expansion while reinforcing AWS-native machine learning as a strategic enabler.