An Agentic AI System for Automated Data Analysis and Machine Learning
Neeru Neeru · Theseus (Ammattikorkeakoulujen) · 2026
The purpose of this thesis was to build a free and easy to use agentic AI web application that automatically does the complete data analysis and machine learning process for any user without needing any technical knowledge. Most existing data analysis tools are too expensive, too complicated, and require strong technical skills that everyday users do not have and this thesis addressed that problem. The thesis was done for self-development and academic purposes at HAMK University of Applied Sciences. The theory part covers the main concepts of agentic AI, multi agent systems, automated machine learning, and explainable AI. The system was built using Python and FastAPI for the backend and Node.js for the frontend, and LangChain and LangGraph for building and connecting the AI agents. The main development method was iterative prototyping where each part of the system was built and tested separately before everything was connected into one complete working application. The system has five agents working together in a LangGraph pipeline. The data agent cleans the data, the model agent trains and selects the best model, the file agent handles CSV, PDF, and image files, and the orchestrator manages and coordinates all the agents. The system was tested on three publicly available datasets which were the Titanic dataset, the Wine Quality dataset, and the India Hockey Match Data dataset as well as PDF and image file. The results showed that the system successfully completed the full automated pipeline for all three datasets without any errors or manual involvement. The Titanic dataset achieved 95.1% cross-validated accuracy, the Wine Quality dataset achieved 92.2% cross-validated R², and the India Hockey Match Data achieved 99.9% cross-validated R² and all three models were automatically deployed above the 82% confidence threshold. The complete source code is published on GitHub and freely available for anyone to use or improve. The source code repository is available at: https://github.com/NeeruSisodia/AgentML-Studio Keywords Agentic AI, multi agent systems, automated machine learning, LangChain, LangGraph, data analysis automation, explainable AI Pages 43 pages and appendices 2 pages