MIT IAP NANDA - AI Agent Development Course

5-Day Intensive Course on Building and Deploying AI Agents

Welcome to the MIT IAP NANDA course! Over 5 days, you'll progress from building a simple AI agent to deploying a sophisticated agent capable of competing in an agent battle.

Course Structure

Day 1: Agent Loop + AI Twin v0

What You'll Build: A simple agent loop (max 5 turns) and your first AI twin using CrewAI (no tools/memory yet)

Key Concepts:

• Understanding agent loops
• CrewAI fundamentals
• Agent, Task, and Crew concepts
• GitHub repository setup
• Creating AgentFacts for your agent

Prerequisites: Python 3.10+, OpenAI API key, GitHub account

---

Day 2: Memory + MCP Tools

What You'll Build: Enhance your agent with memory capabilities and MCP (Model Context Protocol) tools

Key Concepts:

• Short-term and long-term memory
• MCP tool integration
• Using external APIs (e.g., Spotify, web search)
• Tool selection and usage patterns

Prerequisites: API keys for chosen MCP server (e.g., Spotify, Weather, etc.)

---

Day 3: Deploy on Railway + REST API

What You'll Build: Deploy your agent on Railway and expose it via REST API

Key Concepts:

• FastAPI integration
• Cloud deployment on Railway
• REST API design for agents
• Testing in NANDA testbed
• Environment variable management in production

Prerequisites: Railway account

Bonus: Chat Frontend - A simple Next.js chat interface where students can share and chat with each other's agents!

---

Day 4: A2A Communication Protocol

What You'll Build: Agent-to-Agent (A2A) communication system

Key Concepts:

• Agent-to-agent (A2A) communication protocol
• Message routing and agent discovery
• AgentFacts for capability sharing
• Cross-agent collaboration
• Central registry integration

Prerequisites: Working agents (local or deployed)

---

Day 5: Coordination Protocol + Agent Battle

What You'll Build: Advanced coordination and compete in the final agent battle

Key Concepts:

• Agent coordination protocols
• Multi-agent task decomposition
• Agent optimization strategies
• Using advanced tools (web search, stock data, specialized APIs)
• Memory optimization
• Response speed vs. accuracy tradeoffs

The Challenge: Build the most capable agent possible! Agents will be tested and evaluated using the Agent Smart Score system that measures:

• Accuracy and correctness
• Response speed and efficiency
• Reasoning quality
• Robustness across diverse topics
• Collaboration capabilities

Prerequisites: Deployed endpoint + consistent input/output format

---

Getting Started

1. Clone this repository:

   git clone https://github.com/projnanda/5-day-course.git
   cd 5-day-course

2. Start with Day 1:

   cd day-1

   Follow the README in each day's folder!

3. Progress through each day at your own pace, or follow along with the course schedule.

4. Optional: Use the Full Chat System (after Day 3):

   See interface folder for complete instructions!

   Quick version:

   cd interface
   npm install && npm run dev

   Now students can share and chat with each other's agents!

Resources

• CrewAI Documentation
• Model Context Protocol Servers
• Railway Deployment Guide

Course Goals

By the end of this course, you will:

• Understand AI agent architectures
• Build agents with memory and tools
• Deploy production-ready agent APIs
• Implement multi-agent coordination and communication protocols
• Compete in the Agent Battle with Agent Smart Score evaluation
• Have a portfolio-worthy project on GitHub

Tips for Success

1. Start early each day - Building agents takes time!
2. Experiment freely - Try different approaches
3. Ask questions - Your instructors and classmates are here to help
4. Test thoroughly - Agents can behave unexpectedly
5. Have fun - AI agents are exciting technology!

License

MIT License - Feel free to use this for learning!

---

Built for MIT IAP 2026

Powered by CrewAI and NANDA