Gurugram: Ethara AI, an Indian artificial intelligence research and infrastructure company co-founded by Mahanaaryaman Scindia, is expanding its work in reinforcement learning and agentic AI through collaborations with Google Cloud and Amazon Web Services (AWS), as it looks to build training and evaluation infrastructure for increasingly autonomous AI systems.
Based in Gurugram, Ethara AI focuses on Reinforcement Learning as a Service (RLaaS), data curation, model evaluation and post-training systems. Its work includes Reinforcement Learning from Human Feedback (RLHF), Supervised Fine-Tuning (SFT), reinforcement-learning environments and custom evaluation loops designed to improve the reliability and alignment of large language models.
The company was co-founded by Mahanaaryaman Scindia and Suryansh Rana, while Shubham Garg serves as co-founder and Chief Financial Officer. Rana is co-founder and Chief Executive Officer, while Scindia is co-founder and Chief Growth Officer. Public professional information lists Scindia’s association with the company from January 2023.
Google Cloud supports reinforcement learning workflows
Ethara AI has selected Google Cloud and Google Workspace as part of the technology stack supporting its reinforcement-learning operations. According to a Google Cloud customer case study, the company integrated the Gemini API into its training loops, where Gemini can function as an inference-based reward and judge model to score agent trajectories against defined criteria.
Google Cloud said the deployment has helped Ethara AI achieve 99% pipeline uptime and two-times faster harness development, while creating a more unified workflow for researchers working on reinforcement-learning experiments.
Scindia said the company wanted to avoid assembling its infrastructure from disconnected systems.
“We run complex, iterative reinforcement learning workflows for clients across multiple industries, demanding infrastructure that’s reliable, scalable, and secure. We didn’t want a patchwork of tools. Google Cloud and Workspace gave us a cohesive, enterprise-grade ecosystem that can grow alongside our business.”
— Aaryaman Scindia, Co-founder and Chief Growth Officer, Ethara AI
The comments were published as part of Google Cloud’s case study on Ethara AI.
For an AI company, avoiding a “patchwork of tools” may sound like an unusually mundane objective, but when models are repeatedly trained, evaluated and retrained, the plumbing can matter almost as much as the intelligence flowing through it.
Building what Ethara calls the ‘human intelligence layer’
Ethara AI positions its work around the layer between increasingly capable foundation models and the human feedback, expert evaluation and structured environments required to refine their behaviour.
Its focus includes creating training environments in which AI systems can interact, receive feedback and improve through reinforcement learning rather than relying solely on conventional pre-training.
The company says its research also covers agentic AI frameworks, evaluation infrastructure and specialised data curation for frontier large language models. Ethara’s public material describes its approach as combining RLHF, supervised fine-tuning and human-in-the-loop evaluation.
Such post-training techniques have become increasingly important as developers move beyond AI systems that simply generate responses towards agents expected to reason across several steps, use tools and perform tasks with greater independence.
AWS collaboration focuses on agentic AI
Ethara AI is simultaneously developing its relationship with Amazon Web Services.
Earlier in 2026, Scindia and Rana met AWS executives at the company’s Seattle headquarters to discuss enterprise agentic AI, reinforcement-learning infrastructure and scalable training systems. Ethara subsequently described the engagement as the beginning of a partnership journey with AWS.
Rana has also represented Ethara AI at the AWS Summit Bengaluru 2026, where he discussed the evolution of artificial intelligence from systems focused largely on generating text towards models capable of reasoning and operating more autonomously.
More recently, Scindia met Elizabeth Baker, Vice President, Private Pricing at Amazon Web Services, for discussions centred not only on infrastructure but also on how enterprises can prepare their software engineering teams for greater use of AI.
According to Ethara AI, the conversation included possible collaboration around AI education, workforce transformation and programmes aimed at helping engineers work effectively alongside intelligent systems.
That reflects a broader challenge facing companies adopting generative and agentic AI: installing the technology is one job; making sure employees know what to do with it is another.
Scindia stresses people over technology
Scindia has also increasingly framed Ethara AI’s strategy around building an AI ecosystem rather than focusing solely on computing infrastructure.
In a recent post, he argued that technology by itself does not create successful ecosystems and highlighted the role of founders, researchers, mentors, partners and employees in creating sustainable innovation networks.
“At Ethara AI, we’re not just building technology,” Scindia wrote, adding that the company was also developing relationships, talent, partnerships and a community that believes India can play a meaningful role in shaping artificial intelligence rather than merely consuming it.
The argument fits with Ethara AI’s efforts to work with researchers, universities and specialist contributors, particularly as AI developers seek higher-quality human feedback and increasingly specialised expertise for model training.
India’s place in the frontier AI race
Ethara AI’s expansion comes as India’s role in artificial intelligence is shifting from being primarily a market for global AI products and a source of engineering talent towards becoming a base for model development, AI infrastructure, evaluation and post-training research.
The Gurugram company is betting that reinforcement learning will become an increasingly important part of that transition.
Its premise is straightforward: as foundation models become more capable, improvement will depend not only on feeding them more information but also on creating better environments in which they can act, make mistakes, receive feedback and learn.
Ethara AI’s collaborations with Google Cloud and AWS therefore represent more than a cloud-computing decision. They illustrate the increasingly complex infrastructure required to develop agentic AI — where computing capacity, evaluation systems, human expertise and workforce skills have to work together.
For India’s growing AI ecosystem, the larger test will be whether companies such as Ethara AI can convert that combination of talent, infrastructure and global technology partnerships into research and products that compete beyond the domestic market. The race for smarter AI models is already crowded; building the systems that teach those models how to behave may prove just as important.
