Avtomatizacija raziskovalnega procesa z agentnimi velikimi jezikovnimi modeli

Scientific discovery requires both creative hypothesis generation and rigorous experimental validation, processes that have historically demanded human expertise. Large Language Models (LLMs) demonstrate sophisticated reasoning and code generation capabilities, yet previous automation efforts have mostly addressed only isolated research tasks such as literature review or data analysis rather than the complete scientific process. The fundamental question remains whether computational systems can independently generate novel research hypotheses and execute experimental validation without human guidance. Here we show that ten specialized LLM agents operating across six sequential stages can complete entire research experiments from initial conception through experimental implementation to scientific paper production. The pipeline, tested in the domain of multi LLM agent Systems, generated multiple novel research contributions achieving performance improvements of up to 5% over baseline methods while in some cases reducing computational costs, with results validated by domain researchers as methodologically sound and novel. While most previous systems required human-provided research questions or pre-defined experimental frameworks, our pipeline independently identifies research opportunities and designs appropriate validations, demonstrating that LLMs can automate hypothesis generation and testing. These findings, along with notable related work, indicate that barriers to scientific automation are now coordination challenges rather than reasoning limitations.