Project Roadmap: Smart Campus Helpdesk with Agentic AI

PHASE 1: Requirements & Setup

Goals:

• Agents communicate with each other (MCP style)
• Run locally using Ollama or LM Studio (offline LLMs)
• Users interact via a chat interface
• Agents pull data from a real database
• Modular, so more agents can be added later

Tools Needed:

• Python: Agent logic & backend
• FastAPI or Flask: API backend
• SQLite or MongoDB: Database
• Ollama / LM Studio: Running LLMs locally
• React.js or HTML/JS: Chat frontend
• LangChain (optional): Agent chaining
• Postman: For API testing
• VSCode: Development

PHASE 2: Agent Design

Agents:

1. OrchestratorAgent - Routes queries to correct agent
2. AcademicAgent - Handles class schedules, subjects, exams
3. FinanceAgent - Handles fees, payments, due dates
4. EventAgent - Handles events, workshops, fests

MCP-style Message Format (JSON):

{
  "sender": "OrchestratorAgent",
  "receiver": "AcademicAgent",
  "action": "get_exam_schedule",
  "data": {"student_id": "CU2025AI001"}
}

PHASE 3: Database Design

Use SQLite or MongoDB.

Example Tables:

• courses(id, name, instructor, schedule)
• fees(student_id, amount_due, due_date)
• events(id, title, date, description)
• users(student_id, name, email)

PHASE 4: Backend + Agent Logic

Use FastAPI with endpoints:

• /chat (POST) → OrchestratorAgent
• /academic → AcademicAgent
• /finance → FinanceAgent
• /events → EventAgent

Each agent:

• Parses MCP messages
• Queries DB
• Returns JSON response

PHASE 5: LLM Integration (Ollama / LM Studio)

Use Ollama’s HTTP API with models like mistral, gemma.

Example:

import requests

def call_ollama(prompt):
    response = requests.post("http://localhost:11434/api/generate", json={
        "model": "mistral",
        "prompt": prompt
    })
    return response.json()["response"]

PHASE 6: Chat Interface

Tech Stack:

• React.js or HTML/JS

UI Components:

• TextBox for input
• Message bubbles
• Agent avatars
• Display chat log

Flow:

Frontend posts user input to /chat → response displayed in UI

PHASE 7: Communication Flow

1. User types a question
2. Frontend sends to /chat (OrchestratorAgent)
3. Orchestrator uses LLM to decide target agent
4. Sends MCP message to agent
5. Agent queries DB and replies
6. Orchestrator formats response
7. Frontend displays it

PHASE 8: Optional Features

• Voice input via Web Speech API
• Add new agents (e.g., LibraryAgent)
• Save chat history
• User login system
• LangChain for agent chaining

Folder Structure

/smart-campus-helpdesk
├── backend/
│   ├── main.py
│   ├── agents/
│   │   ├── orchestrator.py
│   │   ├── academic_agent.py
│   │   ├── finance_agent.py
│   │   └── event_agent.py
│   ├── database/
│   │   ├── db.py
│   │   └── models.py
│   └── utils/
│       └── ollama.py
├── frontend/
│   └── chat-ui/
├── requirements.txt
└── README.md

Learning Outcomes

• Build and deploy multi-agent architecture
• Route messages using MCP
• Run offline LLMs via Ollama/LM Studio
• Build real-time chat interface
• Master DB + API integration