Joint Industry-Academia Research Unveils Novel Framework on Multi-Agents

The research which has applicability not only for autonomous cloud operations but also AI engineering in general has been shortlisted for the Best Paper award at the International Conference on AI Engineering 2025.

The SA4S (Software Architecture for Sustainability) group at IIITH’s Software Engineering Research Centre has an ongoing partnership with MontyCloud, a CloudOps startup that leverages autonomous bots to manage cloud compliance, security, and continuous operations. When they created an autonomous copilot named Marvin, it functioned as a conversational AI agent allowing users to interact with the platform while simultaneously simplifying everyday cloud operations. In addition to this, embodying its inherent intelligence, the copilot also performed automatic checks, generating actionable reports for cloud users. 

Multi-Agent Architecture
“Marvin was leveraging a single LLM  to answer user queries based on the cloud footprint data,” remarks Prof. Karthik Vaidyanathan, SERC, IIITH. There were various challenges related to maintainability, extensibility and data management. It was then that the group realised the need for an agentic approach. “We thought if multiple agents, each with its own knowledge can collaborate to achieve a specific functionality, we could overcome a number of limitations of our existing system,” continues Prof. Vaidyanathan, referring to the complexity of managing systems on the cloud such as diverse data sources, orchestration of multiple processes and handling complex workflows to automate routine tasks. Initially relying on an retrieval augmented framework – a single LLM combined with an external information retrieval system, the researchers began to explore a multi-agent architecture where each agent could focus on a specific task and bring in different perspectives too. “If you look at any system, you have domain experts, each with their own capabilities possessing domain knowledge like say, in security, performance, sustainability and so on. Similar to multiple people looking at data is the concept of multi-agents that we leveraged here,” says the Professor.  

Tailored Framework
Existing frameworks for the development of multi-agent architectures however did not work out for the SA4S group. “Adhering to pre-existing frameworks did not allow us the flexibility of using particular platforms to power certain agents. We wanted to go a step further where we could decide which agent could be built using which framework. That’s how the MontyCloud team and ours created Meta Orchestrator of Your Agents (MOYA) – where we orchestrate how different agents can be built with different technologies,” explains Prof. Vaidyanathan. Rudra Dhar, PhD student at SERC adds, “While existing multi-agent frameworks are available, they did not meet our specific needs. Developing our own framework, tailored to our use case, but designed to be generic, was a highly rewarding experience, especially since existing solutions fell short in addressing our requirements.”

Sreemaee Akshathala who is a research engineer on the project says, “Developing MOYA exposed us to the challenges of integrating GenAI into real-world CloudOps. What began as a research challenge quickly turned into an engineering problem, and watching MOYA grow from an idea to a fully functional framework has been a very rewarding journey.” For Amey Karan, Dual Degree student from SERC, the highlight of the project has been the fact that from ideation to implementation of the framework, the experience has been incredibly fulfilling. “It has provided me with valuable learning opportunities,” he states. Like Amey, Adyansh Kakran, another Dual Degree student working on the project underscores the great learning experience that he’s had by working closely with industry on real world problems “at the intersection of software engineering and applied AI”.

Accolades and More
This joint research which was published as a paper titled ‘Engineering LLM Powered Multi-Agent Framework for Autonomous Cloud-Ops’ has been shortlisted for the Best Paper award at at the prestigious International Conference on AI Engineering (CAIN). Speaking about the framework, Kannan Parthasarathy, CTO MontyCloud remarks, “By enabling coordinated action from multiple agents, we are empowering the next generation of autonomous applications”. 

In addition to enabling these frameworks for cloud operations, the research team decided to make it generic so as to enable developers to build their own agents. “We have provided an agent portal along with a toolkit where developers can customise and define capabilities for their agents. We are also trying to open source this so that people can create their own use cases,” states Prof. Vaidyanathan. According to him, what’s really unique about this research-industry collaboration is that the framework was developed as a result of real-life challenges that they faced. “Engineering AI systems is always fraught with a lot of challenges. The collaborative research that resulted in the paper publication addresses some aspects of it, and we believe that is an important step.” He adds that a hackathon where users can build use cases using MOYA is being planned on IIITH campus on March 15 in collaboration with the Open Source Developers Group.