A conversational multi-agent application for "MLOps" orchestration
Loading...
Date
2026-03-05
Authors
Μωραΐτης, Ανδρέας
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Advancements in Machine Learning (ML) have created a steady increase in applications across industries. Consequently, a growing demand by organizations for a combination of domain expertise and the use of AI’s computational power and efficiency, requires human-machine collaboration towards more effective decision-making and innovative solutions. In addition, as ML workflows become more complex, MLOps platforms are an essential tool for optimizing the development, deployment and management of these workflows. However, as these platforms offer an extensive set for orchestrating ML pipelines, they can introduce complexities and a steep learning curve for data scientists and users without prior experience on system management and DevOps processes.
Introduced in recent years, Large Language Models (LLMs) can offer a solution to a seamless experience in addressing these challenges based on their capability of understanding and processing natural language as well as utilizing tools to perform actions and communicate with such external services.
This thesis is motivated by collaboration efforts made on the ongoing Horizon Europe project “HumAIne” and its single-agent assistant “Swarm Agent”. It aims to present an alternative design of the LLM based conversational assistant that can access tools and services to manage ML workflows and operations required in the ML life cycle. The implementation follows a multi-agent architecture built with OpenAI’s Agents SDK framework that integrates Kubeflow, an open-source MLOps solution built on top of Kubernetes, enabling ML pipeline orchestration through context-aware guidance from natural language querying. Additionally, a chatbot UI was developed with Chainlit, for a simplified and user friendly conversational experience with the assistant.
Description
Keywords
MLOps, Large language models, Agentic AI, Agents, Chatbots