HAIChart is a human-AI paired visualization system that combines the strengths of both human and AI-driven approaches to generate high-quality visualizations from large datasets. The system uses reinforcement learning to iteratively refine visualization recommendations based on user feedback. It introduces a Monte Carlo Graph Search (MCGS) algorithm to efficiently explore the visualization space and a composite reward function to evaluate visualization quality. Additionally, it incorporates a visualization hints mechanism to actively incorporate user feedback, allowing for progressive refinement of the visualization generation process. The system is designed to address three key challenges: exploring the visualization search space effectively, evaluating the quality of generated visualizations comprehensively, and integrating user feedback to align visualizations with user needs. HAICHART outperforms both human-powered tools (e.g., Tableau, PowerBI) and AI-powered automatic tools (e.g., Draco, Table2Charts) in terms of effectiveness and efficiency. It achieves a 21% improvement in Recall and is 1.8× faster than human-powered tools. In comparison to AI-powered tools, it shows 25.1% and 14.9% improvements in Hit@3 and R10@30, respectively. The system is evaluated using quantitative metrics and user studies, demonstrating its superior performance in generating accurate and relevant visualizations. HAICHART is implemented as a reinforcement learning-based framework that iteratively recommends visualizations by incorporating user feedback. It uses a visualization query graph to represent all possible visualizations for a given dataset, enabling efficient exploration of the visualization space. The system generates visualizations by applying a sequence of operations, such as chart types and x/y-axis configurations, and evaluates their quality using a composite reward function that considers data features, visualization domain knowledge, and user preferences. The system also incorporates a visualization hints selection mechanism to actively gather user feedback and refine the visualization generation process. The system's effectiveness is validated through experiments on two real-world datasets: VizML and KaggleBench. HAICHART demonstrates superior performance in terms of accuracy and efficiency compared to existing visualization tools. It is able to generate high-quality visualizations that align with user preferences, making it a valuable tool for data analysis and visualization. The system's ability to iteratively refine visualizations based on user feedback makes it a promising approach for improving the effectiveness of visualization tools in data analysis.HAIChart is a human-AI paired visualization system that combines the strengths of both human and AI-driven approaches to generate high-quality visualizations from large datasets. The system uses reinforcement learning to iteratively refine visualization recommendations based on user feedback. It introduces a Monte Carlo Graph Search (MCGS) algorithm to efficiently explore the visualization space and a composite reward function to evaluate visualization quality. Additionally, it incorporates a visualization hints mechanism to actively incorporate user feedback, allowing for progressive refinement of the visualization generation process. The system is designed to address three key challenges: exploring the visualization search space effectively, evaluating the quality of generated visualizations comprehensively, and integrating user feedback to align visualizations with user needs. HAICHART outperforms both human-powered tools (e.g., Tableau, PowerBI) and AI-powered automatic tools (e.g., Draco, Table2Charts) in terms of effectiveness and efficiency. It achieves a 21% improvement in Recall and is 1.8× faster than human-powered tools. In comparison to AI-powered tools, it shows 25.1% and 14.9% improvements in Hit@3 and R10@30, respectively. The system is evaluated using quantitative metrics and user studies, demonstrating its superior performance in generating accurate and relevant visualizations. HAICHART is implemented as a reinforcement learning-based framework that iteratively recommends visualizations by incorporating user feedback. It uses a visualization query graph to represent all possible visualizations for a given dataset, enabling efficient exploration of the visualization space. The system generates visualizations by applying a sequence of operations, such as chart types and x/y-axis configurations, and evaluates their quality using a composite reward function that considers data features, visualization domain knowledge, and user preferences. The system also incorporates a visualization hints selection mechanism to actively gather user feedback and refine the visualization generation process. The system's effectiveness is validated through experiments on two real-world datasets: VizML and KaggleBench. HAICHART demonstrates superior performance in terms of accuracy and efficiency compared to existing visualization tools. It is able to generate high-quality visualizations that align with user preferences, making it a valuable tool for data analysis and visualization. The system's ability to iteratively refine visualizations based on user feedback makes it a promising approach for improving the effectiveness of visualization tools in data analysis.