Autonomous Mobility MLOps with AWS Migration
Key Challenges
Teraki faced challenges during their AWS migration, including managing diverse data like camera feeds and 3D models, scaling infrastructure for business growth, and handling scattered data storage. They struggled with variable machine utilization and needed robust MLOps for continuous model training. Additionally, they faced risks in data security, model stability, and potential downtime, with a critical need for effective cost management to optimize their cloud solution's TCO.
Key Results
Teraki's migration to AWS with Ankercloud's collaboration significantly enhanced their operations and data science capabilities. Automated pipelines streamlined model training and deployment, reducing errors and latency. AWS improved data management and security, while cost visibility enabled Teraki to predict and manage expenses effectively. Finally, the new MLOps pipeline empowered customers with self-service model training, increasing engagement and satisfaction. Overall, this transformation aligned perfectly with Teraki's strategic goals for their autonomous mobility services.
Overview
Teraki's mission is to enable personalized and accurate machine learning based applications without a compromise towards 100% safety. They work on automotive and IoT applications for which lower latencies, reduced costs and enhanced accuracy are essential. They were able to scale their ML and MLOPs capability with a successful migration to AWS in collaboration with Ankercloud.
Challenges
Teraki works with three different data categories to offer three key areas of ML/AI services to its customers – camera feed, RADAR and 3D modeling data. For each city it operates in, they create an in-depth 3D model, which is retrained every month. Furthermore, live camera feeds and RADAR data is used to infer environmental objects to help the vehicles and robots navigate itself in an autonomous way. They currently train these models on-prem with distributed data storages across multiple on-prem systems. These machines have high variability in utilization and compared to their business growth, it is difficult to scale. Furthermore, their data science teams is also planned to grow 3x over the next 3 years.
Solution
- Ankercloud setup CI/CD and Terraform based automated ML and MLOps pipeline
- Ankercloud further migrated the existing models and workloads with end-to-end testing, assessment and analysis of their cloud environment to ensure a successful and cost-effective scaling of services on cloud
Business Outcome
- The migration to AWS enabled Teraki to successfully scale their operations and data science capabilities. By implementing automated CI/CD and Terraform pipelines, they facilitated seamless and efficient model training and deployment processes. This automation not only reduced manual efforts but also helped in minimizing errors and latency.
- The move to AWS improved their management of diverse data types and volumes, enhancing data security and access. Teraki achieved better visibility into their TCO per model and per customer, which helped in managing and predicting costs effectively, thereby supporting a scalable and financially viable business model.
- Lastly, the new MLOps pipeline empowered customers to bring their own data and train models specific to their needs, enhancing customer engagement and satisfaction.
- Overall, the AWS migration and collaboration with Ankercloud transformed Teraki’s capabilities, aligning with their strategic goals of safety, accuracy, and scalability in autonomous mobility services.
