Design and Implementation of an AI Agent Web Application for Robotic Process Automation
Sarkar, Arpan (2026)
2026
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202604277955
https://urn.fi/URN:NBN:fi:amk-202604277955
Tiivistelmä
In recent years, automation has played a critical role in improving productivity and reducing manual labour across various industries. Many organizations rely on traditional Robotic Process Automation (RPA) tools to automate repetitive tasks; however, such systems are often limited to rigid, predefined workflows and require significant financial investment and technical expertise. These limitations make traditional RPA solutions difficult to adopt and maintain, particularly for small and medium-sized enterprises (SMEs).
This thesis presents the design and implementation of a web-based AI agent system that enables natural-language-driven robotic process automation for structured data tasks. The developed system allows users to perform operations such as numerical calculations, data cleaning, and data visualization on CSV files through natural language instructions. Google Gemini Pro is utilized as the large language model (LLM), while Microsoft Semantic Kernel is employed to orchestrate reasoning and controlled tool execution. The backend of the system is implemented using FastAPI to mediate between user input, agent reasoning, and deterministic automation tools. Secure authentication and data management are provided through Supabase, and a responsive web-based user interface is developed using React, TypeScript, and Tailwind CSS.
Functional testing demonstrated that the AI agent is capable of correctly interpreting user instructions, processing CSV files, and executing supported automation tasks in a reliable and deterministic manner. While the current implementation focuses on a single RPA use case involving structured CSV data, the system architecture is designed to be extensible and can support additional automation tasks in future development. The results of this work demonstrate the technical feasibility of combining large language models with deterministic automation tools to create flexible and user-friendly RPA systems suitable for practical use in intelligent automation applications.
This thesis presents the design and implementation of a web-based AI agent system that enables natural-language-driven robotic process automation for structured data tasks. The developed system allows users to perform operations such as numerical calculations, data cleaning, and data visualization on CSV files through natural language instructions. Google Gemini Pro is utilized as the large language model (LLM), while Microsoft Semantic Kernel is employed to orchestrate reasoning and controlled tool execution. The backend of the system is implemented using FastAPI to mediate between user input, agent reasoning, and deterministic automation tools. Secure authentication and data management are provided through Supabase, and a responsive web-based user interface is developed using React, TypeScript, and Tailwind CSS.
Functional testing demonstrated that the AI agent is capable of correctly interpreting user instructions, processing CSV files, and executing supported automation tasks in a reliable and deterministic manner. While the current implementation focuses on a single RPA use case involving structured CSV data, the system architecture is designed to be extensible and can support additional automation tasks in future development. The results of this work demonstrate the technical feasibility of combining large language models with deterministic automation tools to create flexible and user-friendly RPA systems suitable for practical use in intelligent automation applications.