FAQs

MAESTRO (stands for Machine Learning & AI Enterprise-level Secure Tool-Suite for Reliable Operations) is a central WOG platform for AI/ML operations. It offers a comprehensive suite of tools and services, along with scalable compute resources, for users to develop, deploy and monitor AI/ML models at scale within a secure environment.

MAESTRO is able to support data up to CONFIDENTIAL (CLOUD-ELIGIBLE), Sensitive HIGH classification (For Sensitive HIGH datasets, it is only applicable for system-ingested datasets from Agency's data systems, and this also requires API integration with Agency's data systems).

MAESTRO is open to all Singapore public service officers with a valid Techpass email account and non-SE GSIB or COMET machine.

The AWS cloud-native tools/services provided by MAESTRO include:

1. Amazon SageMaker Studio (MLOps)
2. Amazon Jumpstart (one click AI/ML model deploy)
3. Amazon SageMaker Canvas (no code AI/ML)
4. Amazon Bedrock (GenAI API services) – Support up to Confidential (Cloud-Eligible)

These are the additional features that MAESTRO has built on top of cloud-native offerings:

1. MAESTRO User Portal - simplified project/user management
2. MLOps Project Templates for R and Python
3. SHIPHATS Integration (Gitlab Runners, Nexus Repository)
4. GGFU Quantised HuggingFace Model Inference Container
5. S3 File Explorer plus Storage for each domain - upload data from your local machine
6. Bespoke Data Integrations with Agency Systems (please raise support ticket) 
7. WOG Service Integrations (Datahive, Data.gov, Cloak, APEX, AI Bots, Ownself Gather, AISAY, OneMap, Fraud Detection Platform, Sentinel)
8. Model Inference Integrations (GEN-Routable API Gateway, CStack, Airbase)
9. Container Image Builder - bring your own docker image
10. Model Telemetry and Cost Utilisation Dashboard
11. MLFlow for ML/LLM traceability and experiment tracking

Aside from the default solution offerings provided by the native cloud service, MAESTRO’s provide a more seamless and holistic AI/MLOps experience for government agencies through various additional engineering enhancements. 

1. Service availability – MAESTRO’s Amazon SageMaker is hosted in GCC's Intranet-facing environment, and additional development and testing has been done to ensure the viability and availability of our services.
2. Diverse AI/ML Capabilities – Besides Amazon SageMaker Jumpstart, which provides users with access to a selection of foundation models and pre-built ML algorithms, MAESTRO also, allows users to host quantised models for greater resource efficiency. In terms of Gen AI offerings, MAESTRO grants access to AWS Bedrock that is exclusively hosted in the Singapore region, providing models such as Claude 3.5 Sonnet & Claude 3 Haiku, while also allowing users to call models from other clouds like Azure via APIs. 
3. WOG AAD / Techpass Authentication – Onboarded users will just have to login via WOG AAD or Techpass authentication instead of SSO login into the main AWS console.
4. User Domain Management Portal – Our custom-built front end allows users manage their projects (members, feature requests, access controls, endpoint creation etc.) via a user-friendly interface.
5. API Gateway for Amazon SageMaker Endpoints – This gateway is provided for MAESTRO users to deploy models via endpoints, from our Sagemaker environment to their respective agency’s destination systems.
6. Security and Permissions configurations – These are already pre-set within MAESTRO’s Amazon SageMaker service according to IM8 security and compliance requirements.
7. R and Python Project Templates – Our product team built these custom project templates to allow users to quickly develop on both Python and R within MAESTRO’s Amazon SageMaker environment.
8. Access to up-to-date packages and libraries, along with the means to collaborate and share code repositories amongst project team members.

MAESTRO is currently free to use until further notice.

However, project teams are encouraged to track their resource utilisation via our web portal’s interface to ensure proper accountability.