Title: Experimental Approaches to Agentic AI in Scientific Research

Abstract: This study introduces two innovative frameworks designed to integrate autonomous, agentic AI into scientific processes. Both architectures employ a hybrid model combining local bodies with remote brains, executed via Google Colab. In this setup, Python-based local orchestrators manage interactions with large language model (LLM) cloud services. The first system, named DeepTS/DeepCollector, is engineered to automate the extensive curation, extraction, and deduplication of time-series data. The second system, DeepScribe, functions as an autonomous analyzer for presentations, transforming physics lectures that are visually dense and mathematically complex into structured scientific reports. By implementing practical systems engineering solutions—such as distributed concurrency controls, remote data inspection, and granular attribute extraction through Cellular RAG—we illustrate how agentic AI can bypass the context and reasoning constraints of current state-of-the-art models to provide rigorous support for scientific workflows. Additionally, we propose a generalization of DeepTS to facilitate deep knowledge graphs and explore the potential application of this conceptual framework to high-energy physics, specifically within the domain of DeepQCD.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC