About MokSa.ai
mokSa.ai was founded in 2021 with a vision to use Artificial Intelligence (AI) to address various problems faced by business communities. The company delivers customized video intelligence solutions for all types of business.
The company’s platform detects customer and employee theft in real-time, utilizing sophisticated algorithms to help businesses reduce substantial financial losses due to criminal activity.
Aiming to create a scalable, affordable solution for small businesses with limited budgets and resources, utilising AWS cloud technology, Moksa’s CTO, Kranthi Mottu, approached the hyperscaler’s French division for partner recommendations. He was recommended Automat-it, an AWS Premier Tier Partner, for their expertise with the AWS Migration Acceleration Program (MAP).
The Challenge
Kranthi Mottu aimed to upgrade MokSa.ai’s cloud architecture, which initially relied on EC2 without containers and local storage, creating scalability limitations. Initially, the solution required each model to run on a dedicated GPU at runtime, needing GPU instances per customer. This setup led to underutilized GPU resources and elevated operational costs. The goal was to transition to a more scalable AWS-based setup to improve performance, support a broader rollout, and address key challenges:
- Data migration: move surveillance data to the cloud, ensuring data integrity and security.
- Scalability: support increased data loads and user growth.
- AWS expertise: access flexible, on-demand AWS expertise to accelerate development.
- Development speed: expedite progress with skilled AWS DevOps professionals.
- Employee analysis: add features to analyze and score employee performance.
The Solution
We aimed to create a solution that would limit AWS infrastructure costs to $30 per camera per month while keeping the total processing time under 500 milliseconds so that each mokSa.ai customer received a quality service.
Migration to the Cloud
Through AWS’s Migration Acceleration Program (MAP), MokSa.ai collaborated with Automat-it’s experts to migrate its infrastructure. Automat-it’s Dedicated Engineer Service then provided ongoing support to streamline customer onboarding.
The initial approach
We initially deployed YOLOv8 using a client-server architecture that offloaded preprocessing and postprocessing to CPU instances while reserving GPUs solely for inference, aiming to reduce costs and maintain performance. While the inference time was low (7.9 ms), the network communication overhead (10.26 ms) increased processing time. Although processing time was still acceptable, we knew the costs were not.
A better solution
We further optimized our solution to ensure costs-per-camera were reduced. That meant moving away from a client-server approach and implementing GPU time-slicing in the EKS cluster using the NVIDIA Kubernetes device plugin so AI Models could share a single GPU. This simplified the scaling process and reduced operational overheads. Plus, without needing to maintain custom code, implementation and maintenance were streamlined.
By testing efficiency and performance across three stages we made sure cost reduction was aligned with our benchmarks for service quality. More detail on this process can be found here.
Automat-it’s project milestones
- Built EKS infrastructure, including networking, security, and monitoring
- Implemented Kinesis Video Streams for real-time video data streaming
- Enabled fractional GPUs on EKS to improve ML inference performance and reduce costs
- Developed CI/CD pipelines using GitHub, Jenkins, and ArgoCD
- Supported YoloV8 model training on SageMaker
- Onboarded customers and created workflows to expedite the process
- Established multi-region Disaster Recovery
- Provided technical documentation and per-camera cost estimates
The Results
Within three months, Automat-it completed MokSa.ai’s migration to an AWS-powered, scalable, and cost-effective platform. By optimizing AI-driven features like predictive analytics and computer vision, they enhanced the platform’s performance, enabling real-time theft detection via EKS-managed GPU instances.
The test results demonstrated that GPU time-slicing enabled the maximum number of AI models to operate efficiently on a single GPU, significantly reducing costs while providing high performance.
- GPU usage was reduced to $27.31 per month, per camera – a twelvefold reduction
- Performance of the YOLOV8-based AI models was maintained in appropriate service range
- Minimal maintenance and modifications to the model code were required with this method – unlocking scalability
Automat-it’s DevOps as a Service model also enabled MokSa.ai to rapidly scale its DevOps team without the high costs and time demands associated with recruiting, thereby maintaining project momentum. This collaboration also facilitated the seamless integration of MokSa.ai’s platform with customers’ existing cameras, making advanced surveillance accessible to small businesses without requiring costly hardware upgrades.
Customer testimonial
Kranthi Mottu comments:: “The support from Automat-it’s team overseen by their DevOps Team Leader, Oleksii Korostelov and Senior MLOps Engineer, Vladyslav Melnyk was invaluable in helping us achieve our project goals. Their support and dedication were crucial in transitioning to a scalable cloud-based solution and leveraging AWS’s advanced AI technologies.”
„I recommend Automat-it’s Dedicated Engineer Service to any startup needing AWS expertise without the hassle of recruiting and employing in-house. It’s the perfect solution for saving time, resources, and budget while receiving top-tier DevOps support.„
“Our account manager, Camille Jubin, has been a cornerstone of our success with Automat-it and AWS. Her proactive approach and attention to detail have made a significant impact on our project’s management.”
Achieve a cloud-based solution
MokSa.ai’s journey from a personal experience to developing a groundbreaking workplace surveillance solution highlights the power of innovation driven by passion and commitment to safety. With Automat-it’s support, MokSa.ai has successfully transitioned to a cloud-based solution, integrated advanced AI technologies, and continues to push the boundaries in the surveillance industry.
The company’s dedication to protecting small business employees and reducing losses underscores the broader social impact and importance of its technology.
With its sights now set on supporting organizations in transportation, schools, and universities by providing them with its cutting-edge surveillance platform, MokSa.ai is setting the stage for a safer, more secure future.
Ready to achieve the same results? Get in touch to work with Automat-it.
For a more in-depth, technical look at the collaboration between mokSa.ai and Automat-it read the AWS blog.
