Monte Carlo simulations are essential for replicating real-world scenarios in many scientific fields. However, these simulations, especially those using FLUKA, a versatile Monte Carlo simulation package, can be complex, time-consuming, and prone to errors due to limitations in automation and integration with external tools. Researchers have developed AutoFLUKA, an AI-powered framework that leverages large language models (LLMs) to automate these complex workflows. AutoFLUKA minimizes human intervention and the potential for errors by automating input generation, simulation execution, and even post-processing of results. This innovative approach involves using custom tools within AutoFLUKA to handle various tasks. For instance, AutoFLUKA can modify FLUKA input files according to user-defined parameters, automatically execute multiple simulation cycles, and decrypt the resulting binary output files for easier analysis. A key innovation of AutoFLUKA is its JSON-based data output, which makes it much easier to analyze the results compared to traditional manual methods. Beyond merely executing simulations, AutoFLUKA incorporates a Retrieval Augmented Generation (RAG) tool that acts as a virtual assistant for users. This tool can answer questions, provide guidance on tackling common FLUKA challenges, and offer context-specific help, significantly improving the user experience. AutoFLUKA also handles the crucial aspect of uncertainty quantification in simulations. It automatically calculates the average uncertainty of the results and, if it's above a user-defined threshold, intelligently determines the necessary number of primary particles to simulate in order to achieve the desired accuracy. This automation extends to data visualization. After running the simulations and processing the data, AutoFLUKA can generate plots of the results, saving them as images for immediate review and analysis. The framework is being applied to diverse fields including microdosimetry, where it's used to design and optimize radiation detectors. By automating complex tasks and providing expert assistance, AutoFLUKA empowers researchers to conduct simulations more efficiently and effectively, accelerating scientific discovery and innovation. Future development will focus on enhancing the RAG tool to handle figures and graphs within technical documents and improving the multi-agent framework for even more complex automation. The potential for AutoFLUKA to be adapted for other simulation software with text-based input files opens exciting possibilities across various scientific domains.