Loka Improves Training for Edtech Company’s Speech-Recognizing GenAI Platform
Industry
Apps, EdTech, AI,
E-Learning, Education
Tech & TOOLS
Amazon SageMaker, AWS Lambdas, AWS Fsx Lustre, Docker, AWS ECS, AWS S3, AWS Cloudmap, AWS EC2 + GPU, AWS Elasticache, MongoDB, AWS Eventbridge, and more
Teams & Services
DevOps, ML Engineering,
Project Management
milestones
Successful migration from
GCP to AWS in six months,
Unblocked key ML initiatives
Industry
Apps, EdTech, AI,
E-Learning, Education
Tech & TOOLS
Amazon SageMaker, AWS Lambdas, AWS Fsx Lustre, Docker, AWS ECS, AWS S3, AWS Cloudmap, AWS EC2 + GPU, AWS Elasticache, MongoDB, AWS Eventbridge, and more
Teams & Services
DevOps, ML Engineering,
Project Management
milestones
Successful migration from GCP to AWS in 6 months Unblocked key ML initiatives
65% of U.S. 4th graders are behind a grade level in reading. With Ello, children are reading more than ever before.
The situation
Building a 1:1 tutor was one of the original dreams of AI. It’s finally possible. Ello is combining generative AI, machine learning, education and psychology into an app-based virtual reading coach for kids. The company’s ultimate goal is eliminate childhood illiteracy by harnessing GenAI to tailor education to each specific learner--and to do it at scale. The edtech leader tapped Loka to help migrate from Google Cloud, modernize on AWS and unblock key machine learning initiatives.
The challenge
Ello engaged Loka to assess the workloads, including generative AI functions and tools for speech recognition, to be migrated and modernized to AWS cloud. There were two main drivers for this migration and modernization:
Scalability: To address their increasing need for GPU resources, Ello needs to update from low capacity on Google Cloud.
Availability: To ensure reliable service delivery to their users, they must maintain a high level of availability and meet uptime service level agreements (SLAs).
The solution
Loka identified several needs for Ello in the discovery phase, including control over resource creation and access, implementing infrastructure as code (IaC), optimizing costs for compute workloads and storage, acquiring additional GPU capacity, scaling ML models and ensuring proper security.
Loka concentrated on migrating Ello's architecture from Google Cloud Platform to AWS, emphasizing security, monitoring, GPU capacity and best practices for Ello's environment.
Our migration strategy was to re-platform all GCP services with those that were similar in AWS, refactoring some applications toward new services in AWS to improve scalability, deployment and management.
What we delivered
Test plan: We implemented tests for function, integration, performance and security, ensuring the validity of the migration process.
DevOps optimization: All workloads were constructed using IaC and followed Ello's Continuous Integration/Continuous Deployment (CI/CD) and Software Development Life Cycle (SDLC) policies.
Cutover and rollback plan: We put a safe cutover and rollback plan in place for relevant production workloads in scope or parallel operation of environments until appropriately tested and validated.
DR (Disaster Recovery) and BCP (Business Continuity Plan): The key was to ensure the implementation of Disaster Recovery for all AWS services and BPCs.
Project highlights
Post-migration, Ello experienced a 30-50% increase in GPU resource availability. Additionally, the successful architecture led to faster multiprocessing jobs using CPUs, contributing to an accelerated product development cycle and reduced time to market.
Fast, efficient migration
With Loka's involvement, Ello successfully migrated from GCP to AWS in 50% less time than their estimated timeline without Loka. This significant efficiency is due to Loka's specialized expertise and focused approach, complementing Ello's ongoing machine learning development.
Efficiency gains with FSx Lustre integration
By integrating FSx Lustre, Loka enabled Ello to avoid the drawbacks of handling a 20TB dataset solely through S3, such as the need for full downloads or bandwidth limitations. This strategic move positions Ello to enhance the efficiency and cost-effectiveness of their dataset training processes.
"Loka came in and helped us do two things at once. One, move a lot of our machine learning workflows from GCP onto AWS very quickly. And then in the process of doing that, we modernized our DevOps stack substantially."
