Samba Safety helps reduce driver risk and promote safer communities through data-driven insights. Their software supports automotive and shipping industries by maintaining low-risk driver profiles to prevent accidents and damage.
Challenge
Samba Safety’s data science team built powerful models to enhance driver risk analysis, but their workflow relied heavily on manual processes. Scripts for data preprocessing, model training, tuning, and validation had to be run manually whenever new data arrived, with no automated versioning or hosting for inference.
This manual model promotion process was time-consuming, labor-intensive, and slowed innovation. Samba needed an automated solution that replicated their manual workflow, integrated with their existing external code repository, and supported continuous delivery for both training and inference.
Solution
Firemind designed two ML pipelines using AWS: one for model training and another for inference. The training pipeline ran Samba’s custom preprocessing and training scripts, deploying the resulting model to Amazon SageMaker for both batch and real-time inference. The inference pipeline handled prediction requests separately, ensuring flexibility and scalability.
AWS Step Functions, SageMaker Processing, and AWS Lambda were used to replicate Samba’s manual workflow while adding full automation for training, tuning, deployment, and inference workloads. AWS CodeStar was integrated with Samba’s existing code repository to trigger automated pipelines whenever code updates were committed.
Services used
- AWS Step Functions
- AWS Lambda
- AWS CodeStar
The Results
- 70% reduction in time to delivery
- Faster turnaround times for model deployment
- Improved accuracy in model outputs
- Fully automated ML workflow
- Scalable infrastructure supporting continuous delivery
