# Design and Evaluation of a Multi-Agent Chatbot for Academic Regulations

## Overview

This project presents a **multi-agent chatbot system** designed to interpret and explain university regulations through both **deterministic (rule-based)** and **generative (LLM-based)** reasoning.  
Developed as part of an MSc dissertation at **CITY College, University of York Europe Campus**, the system integrates multiple autonomous agents under a Python-based orchestration layer.

The chatbot aims to make university rules transparent, explainable, and easily accessible to students and academic staff, reducing administrative workload and improving academic guidance.

---

## System Architecture

The system follows a **multi-agent design**, where each agent performs a specific function and communicates through structured data formats (`.txt`, `.csv`).  
An **orchestrator** manages the entire workflow—ensuring each stage executes sequentially, validating outputs, and coordinating agent dependencies.

### Agent Map

| Agent | Description | Key Output |
|--------|--------------|------------|
| **Agent 1 – Regulation-to-Rules Converter** | Converts official university regulation documents (UCC & PCC) into structured IF–THEN rules using OpenAI’s GPT-4o model. | `if_then_rules.txt` |
| **Agent 2 – Transcript-to-CSV Converter** | Extracts module and grade data from `.docx` or `.pdf` transcripts and converts them into standardized CSVs for analysis. | `student_transcript.csv` |
| **Agent 3 – Rule-Based Reasoning Engine** | Applies the encoded IF–THEN rules to student data to determine progression, reassessment, compensation, and award classification. | Interactive terminal results |
| **Agent 4 – LLM Chatbot (GPT-4o)** | Provides natural-language explanations of university regulations and decisions, using either a CLI or Streamlit web interface. | Conversation responses |
| **Agent 5 – Streamlit Interface of LLM Chatbot (GPT-4o)** | Provides natural-language explanations of university regulations and decisions| Conversation responses |