About DopikAI
Established in 2019, DopikAI Is An Artificial Intelligence (AI) Company Specializing In Providing Digital Conversion Solutions In The Field Of Image Processing, Computer Vision, NLP.
One of their solutions is Text/ Image recognition such as Legal Documents Recognition and ID Card Recognition. The challenge is how to build a MLOps pipeline for labeling, training, and deploying models in operation effectively and cost saving.
SotaTek has applied AWS MLOps best practice to solve all challenges
Amazon Simple Storage Service (Amazon S3) is used to store raw image data because it provides us with a low-cost storage solution.
The coordination of the labeling process is managed through AWS Step Functions, a serverless workflow engine designed to streamline the various stages of labeling. Within this process, Amazon SageMaker Ground Truth is leveraged for automated labeling via labeling jobs and managed human resources. Additionally, AWS Lambda is employed for data preparation, initiation of labeling tasks, and storage of labels within Amazon SageMaker Feature Store.
The process of constructing and training models is coordinated through Amazon SageMaker Pipelines, which seamlessly integrates with essential SageMaker functionalities through its built-in steps. Model training is automated using SageMaker Training jobs, while data preparation and model performance evaluation are facilitated through SageMaker Processing jobs.
AWS Codebuild is invoked by AWS Step function to build Docker image and trigger GitOps for automatically releasing a new version of the model to DEV/UAt/PROD environment.
AWS EKS is used for hosting Web applications, With AWS we can easily apply GitOps, Auto Scaling for Saving Cost.
