Agentic AI – Email Classification Using XGBoost, Ollama (gemma3:1b) with Interpretable Dashboard

Agentic AI - Email Classification Using XGBoost, Ollama (gemma3:1b) with Interpretable Dashboard

C. T. M Praveen Kumar Assistant Proffesor(Artificial Intelligence

and Machine Learning) Ballari Institute of Technology and

Management Ballari, Karnataka, India ctm.praveen@bitm.edu.in

Shravani. H. G B.E(Artificial Intelligence and Machine

Learning)

Ballari Institute of Technology and Management

Ballari, Karnataka, India shravanirajulu666@gmail.com

Sunidhi. R. Kulkarni B.E(Artificial Intelligence and Machine

Learning)

Ballari Institute of Technology and Management

Ballari, Karnataka, India sunidhirk05@gmail.com

Yashasvi. S. Kotian B.E(Artificial Intelligence and Machine

Learning)

Ballari Institute of Technology and Management

Ballari, Karnataka, India yashasviskotian@gmail.

Rohith. B. S B.E(Artificial Intelligence and Machine

Learning)

Ballari Institute of Technology and Management

Ballari, Karnataka, India rohithbs245@gmail.com

Abstract—The Agentic AI–based Phishing Email Detector with Explainable Dashboard is important as it elevates cybersecurity by safeguarding the users from phishing attacks in real time. Divergent from traditional filters, it furnishes lucid explainations for every decision helping users have faith in the system. With SHAP and ollama it evolves to new hazards seamlessly, while gmail style interface makes it clear and intuitive. This assures both security and conciousness for everyday email users. The systems precision relies on the quality of training data, and emerging phishing techniques may bypass detection. It may produce wrong alerts and missed threats which may impact usertrust. Despite the fact dashboard provides the clarification and explainations, they may still be complicated for the non tech students users. Handling confidential inbox data also triggers confidential concerns. Running AI models on restricted devices can develop efficiency issues. The key purpose of this work is to develop the machine learning based phishing email detection with high accuracy. In the further process it targets to provide explainable insights using SHAP and to provide interactive streamlit dashboard tht allows users to take the input from the email content and view the result and to will fetch real time emails.To incorporate multi-agent framework with ollama for deploying agentic AI facilitating dynamic reasoniing in phishing identification. The project uses the agile development approach, it fetches email using Gmail IMAP and it collects email data from Gmail IMAP for detecting and it will preprocess to clean and extract the features. And also it will help to combine modules into a streamlit dashboard for real-time detection and visualization. Inclusion to this we will add up the agentic AI framework which comprises the multi - agent and olama will make our system automated, Dynamic and interpretable. System will detect phishing emails with high accuracy while minimizing wrongly flagged emails, for each and every prediction it is paid with the ollama explaination and it even helps non-technical background users to understand the threats. In the comparison of the traditional spam filters, even though it provides high accuracy and adaptability, some flagged emails manage to get into the primary inbox. It helps in the clear visulization by providing dashboard, the system guarantees modifiable, self-directed and explainable operations.

Keywords— Email detector, Ollama (gemma3:1b), SHAP, Explainable Dashboard.