Reducing Labeling and Inference Cost of SageMaker
for Construction-Tech Company
Overview
Sphere along with our AI and ML partners, worked with a construction-tech company over a few weeks to comprehensively optimize their SageMaker installation.
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
Our 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.
