Sanas.ai specializes in accent translation and voice enhancement for call centers, aiming to improve communication and understanding between individuals. The company recently secured an investment of $50 million to accelerate its growth and expand its operations.
As part of this growth, the AI team at Sanas.ai faced the challenge of scaling beyond its current size of 7 people.
This case study explores how the team, with the assistance of a group of 3 full-time consultants, efficiently scaled their AI operations and continually improved their MLOPs capabilities to meet the demands of a rapidly growing business.
As Sanas.ai gained traction and more clients, the AI team quickly realized the need to scale their operations. They needed to enhance their technology, improve their development processes, and optimize the entire machine learning workflow to handle a higher volume of requests and data.
To address these challenges, the AI team sought external guidance and expertise, and that's when they engaged our team of 3 experienced consultants.
The first step in scaling the AI team was to expand its size beyond the initial 7 members. Eventum collaborated with the existing team to assess their strengths and weaknesses and identified the key areas where additional talent was required.
Eventum recruited talented ML engineers and data scientists to fill the identified gaps in needed skills.
To ensure smooth collaboration and maintain consistency across the growing team, we introduced a set of best practices. The team adopted technologies and tools such as pytorch_lightning, hydra, unit testing, wandb, poetry, linters, formatters, and typecheckers to streamline their development processes.
The adoption of best practices contributed to a more efficient and reliable development workflow, reducing errors and minimizing the time spent on debugging.
To meet the increasing computational demands, the AI team needed to enhance their infrastructure and optimize their CI/CD pipeline.
Our consultants recommended setting up CI/CD on GPUs and Windows machines which reflected the production environment to speed up the development and testing process. This allowed the team to validate changes faster and catch potential issues early in the development cycle.
Eventum also worked with Sanas.ai to adopt an autoscaling Kubernetes cluster to handle the dynamic workload of machine learning tasks. This elastic infrastructure enabled the team to automatically scale resources based on demand, ensuring efficient resource utilization during peak periods.
With our teams’ assistance, Sanas.ai implemented cost-saving strategies to make their AI operations more economically sustainable.
The team enabled spot instances, taking advantage of unused cloud resources to save up to 70% on training costs. By utilizing spot instances for non-time-sensitive tasks, they maximized their cloud spending efficiency.
Furthermore, the AI team migrated their operations to the AWS cloud, leveraging its robust and scalable infrastructure. The migration enabled the team to benefit from AWS's vast array of services and resources, providing them with the flexibility to adapt to future business growth.
To streamline the development process further, we introduced automatic deployment to production using git tags. This streamlined the deployment process, reducing the risk of manual errors and facilitating seamless rollouts of new features and improvements.
The team also adopted Dockerized repositories, enabling them to package their applications and dependencies consistently. This containerization approach simplified deployment and ensured a consistent environment across various stages of the development lifecycle.
For Extract, Transform, Load (ETL) processes, we recommended using dagster, making it easier for the team to manage data workflows. Sanas.ai became one of dagster's first official Trusted ML Partners, showcasing their commitment to excellence in data engineering and data management.
One of the critical achievements that emerged from the collaboration with Eventum was the implementation of Automatic Merge Request development environments.
This approach allowed developers to run any ETL step starting from any point in the graph automatically. This significant improvement sped up development drastically, as developers could focus on specific parts of the code without waiting for the entire workflow to complete.
Scaling an AI team beyond its initial size is a challenging endeavor, but with the right expertise and best practices, it can be achieved efficiently.
Sanas.ai, with the help of a team of 3 full-time AI specialists, successfully expanded their AI team, improved their MLOPs capabilities, and optimized their infrastructure to support their accent translation and voice enhancement services.
By adopting best practices, enhancing their CI/CD pipeline, optimizing costs, and leveraging cloud resources effectively, Sanas.ai positioned itself for continued growth and success in the dynamic and competitive AI industry.
The Automatic Merge Request development environments became a game-changer, accelerating the team's development process and ensuring they stay at the forefront of innovation in the field of AI-driven communication solutions.
See how Eventum helped launch an app into the App Store in 4 months, currently going viral Generative AI next gen social media.Read the Case Study on Plai Labs
Learn how Eventum helped DeepCell reduce manual oversight by 90%, provided R&D roadmapping & mentorship of ML teamRead the Case Study on DeepCell
Learn how to hire your AI team, ranging from role types & expectations, matching positions to project requirements, and interview structures.Read the White Paper
See how Eventum helped launch an app into the App Store in 4 months, currently going viral Generative Al next gen social media.
Learn how Eventum helped DeepCell reduce manual oversight by 90%, provided R&D roadmapping & mentorship of ML team