Multi-agent AI system automates ML pipeline generation with self-healing

A team of researchers has introduced a comprehensive multi-agent framework designed to automate the creation of end-to-end machine learning pipelines based on datasets and objectives expressed in natural language. This system comprises five agents responsible for profiling, intent interpretation, microservice suggestions, constructing Directed Acyclic Graphs (DAG), and executing tasks. It employs code-grounded Retrieval-Augmented Generation (RAG) for better microservice comprehension, features an explainable hybrid recommendation system, and incorporates a self-healing capability that utilizes LLM-based error analysis and adapts from execution history. When tested across 150 ML tasks, the framework demonstrated an 84.7% success rate for end-to-end pipelines, surpassing existing benchmarks and significantly shortening workflow development time.