The Challenge

This construction-tech company had recently started with SageMaker on AWS, but was not getting the desired cost-basis they desired. This company was also looking to replace their Segmentation model in an effort to reduce infrastructure costs.

How It Was Solved

Gad Benram and his team conducted a comprehensive checkup of the partner’s SageMaker instance through the following project model:

  • ML Checkup: 2-4 hour meetings per week, talking with the teams and mapping the state of ML
  • Develop a Path to ML/Improve Current Metrics: Working with stakeholders, diving into the existing model and building project metrics/KPI’s for success
  • Build, Operate and Transfer: Complete buildout of the ML solution on SageMaker

The Results

Through a comprehensive ML review, this construction-tech company saw the following cost-benefits:

  • Reduced SageMaker inference cost by 40% by updating endpoint configurations
  • Built SageMaker Groundtruth Pipeline to tag 30,000+ images — minimized the need for labor through AI assisted labeling, ultimately reducing the cost of labeling by 70%.
  • Trained SageMaker Bounding Boxes model to replace their Segmentation model—reducing the output size by a factor of 100 and reduced the infrastructure costs for inference jobs.