Challenges and Solution

• AI Costs

  • Challenge: The client initially used AWS Rekognition for AI proctoring, which was cost-effective at a smaller scale. However, as the system scaled, Rekognition became a significant cost driver.

  • Solution: Xponentium conducted a thorough market analysis, comparing hyperscaler and open-source AI options for Image and Video Analysis. A customized open-source solution was identified as the optimal choice, balancing cost with performance and accuracy. Xponentium collaborated with the client’s team and trained the chosen AI model until the required accuracy was met. The migration from Rekognition to the open-source system was executed with a canary approach, ensuring zero downtime.

• EC2 Costs

  • Challenge: Client was using on-demand EC2 instances as a backend server. Most of the compute operations were done these instances, overall it was incurring significant cost. 

  • Solution: Xponentium identified many non-mission-critical tasks, such as report generation and video chunk merging, which were being processed synchronously and were expensive. Then we shifted these workloads to asynchronous processing and moved them to spot instances, significantly lowering costs without impacting user experience or business operations.

• Storage Costs

  • Challenge: The client lacked a clear object lifecycle policy, leading to unnecessary storage of files and images for extended periods. This resulted in higher storage costs without any corresponding benefit.

  • Solution: Xponentium implemented an Object Lifecycle Policy and optimized storage tiering. By analyzing data retrieval patterns and SLAs, Xponentium developed a storage assignment algorithm that efficiently distributed objects across S3 storage tiers, including Standard, Intelligent, and Glacier.