Open Source & Indian AI models in practice

Open-weight models are rapidly maturing, and modern hardware has made local inference increasingly viable. Yet for enterprises, the real challenge is not capability, it is enabling developers to consume diverse models without compromising governance, compliance, cost control, or flexibility.

This session presents a Models-as-a-Service (MaaS) reference architecture built on open-source technologies that provides a governed access layer for enterprise AI.

We demonstrate how organizations can expose local and self-hosted models through a unified API while enforcing authentication, authorization, rate limiting, model routing, audit logging, quota management, and safety guardrails realized through theopen-source MaaS project within the OpenDataHub ecosystem.

Live demo of Sarvam and other india specific models delivered through MaaS in self service and governed way.