The Department of Telecommunication (DoT)’s AI Standardisation Committee released its Indian AI Stack discussion paper, with the intention of mitigating impediments in AI deployment and essentially make AI uniform for application across sectors.

The AI stack will secure storage environments that simplify archiving and extraction from data based on the data classification, ensure the protection of data through data federation, data minimisation, an open algorithm framework, defined data structures, interfaces and protocols, and monitoring, auditing, and logging, as well as ensuring the legitimacy of backend services. 

This paper says the AI stack will comprise of five major horizontal layers and one vertical layer:

• Infrastructure – This would be the core layer that forms the basis of an end to end AI functionality, with a common data controller, private and public cloud setups, determining the manner of data collection
• Storage – This layer outlines the appropriate protocols for storing different kinds of data, for more efficient usage and retrieval depending on the need.
• Compute – This layer facilitates an open algorithm network, with AI/ML processes including NLP, DL and neural networks. Compute layer will also lay emphasis on data engineering, build and deploy service architectures and provide transparent architectures to industry for them to develop their own analytics.
• Application – A “purpose-built” layer that enables software and other applications to be hosted and executed as a service layer, which will also support backend functions and provide frameworks for the AI engine to work smoothly.
• Data Exchange – Also the information exchange layer, this layer will outline the final end user experience through defined structures and protocols. This layer is also responsible for ensuring ethical standards pertaining to digital technologies are followed and adhered to. Until Indian laws are formalised, GDPR laws can be applied.
• Security – This is the sole vertical layer that cuts across the above layers to ensure the AI services are safe and secure.

To understand the application of the layers better, click here for the complete discussion paper.