🧠 Meeting Intelligence Agent

A production-style AI agent that extracts structured insights from meeting notes using schema validation and self-healing.

---

🚀 Overview

This project implements a task-oriented AI agent designed to process meeting text and return structured insights, including:

• Summary
• Action Items
• Decisions
• Risks

The agent is built with:

• OpenAI API
• Strict JSON schema enforcement
• Validation + self-healing retry mechanism
• Modular project structure

---

🎯 Features

✅ Structured Output

Always returns JSON with the following schema:

{
  "summary": "...",
  "action_items": ["..."],
  "decisions": ["..."],
  "risks": ["..."]
}

---

🛡️ Validation Layer

• Ensures correct schema
• Enforces types (string vs list)
• Prevents malformed responses

---

🔁 Self-Healing Mechanism

If the model returns:

• invalid JSON
• incorrect schema

The agent:

1. Detects the issue
2. Sends a repair prompt
3. Retries automatically

---

🎯 Task-Scoped Design

This is not a chatbot.

It is a single-purpose agent: Analyze meeting notes → return structured data

---

📂 Project Structure

meeting-intelligence-agent/
│
├── app/
│   ├── agent.py        # Core agent logic
│   ├── config.py       # Settings (model, debug)
│   ├── prompts.py      # System prompt
│   ├── validators.py   # JSON validation logic
│   └── main.py         # CLI interface
│
├── run.py              # Entry point
├── requirements.txt    # Dependencies
├── .env                # API key (not committed)
├── .gitignore
└── README.md

---

⚙️ Setup

1. Clone the repository

git clone https://github.com/dmartinezbeltrami/meeting-intelligence-agent.git
cd meeting-intelligence-agent

---

2. Create a virtual environment

python -m venv venv

---

3. Activate it

Windows (PowerShell):

.\venv\Scripts\activate

---

4. Install dependencies

pip install -r requirements.txt

---

5. Add your API key

Create a .env file:

OPENAI_API_KEY=your_api_key_here

---

6. Run the agent

python run.py

---

🧪 Demo

Input

We had a meeting with the product team.
John will redesign the signup flow.
Maria will analyze drop-off rates.
There is a risk due to unstable backend services.

---

Output

{
  "summary": "The product team discussed improvements to the signup flow and analytics.",
  "action_items": [
    "John will redesign the signup flow",
    "Maria will analyze drop-off rates"
  ],
  "decisions": [],
  "risks": [
    "Unstable backend services"
  ]
}

This output is automatically validated and repaired if needed.

🧠 Key Concepts Demonstrated

• Prompt engineering for structured outputs
• Schema validation
• Self-healing AI responses
• Task-scoped agent design
• Separation of concerns (modular architecture)

---

🔒 Security

• API keys are stored in .env
• .env is excluded via .gitignore
• No secrets are hardcoded

---

📈 Future Improvements

• Add unit tests for validation
• Add logging system
• Extend to multi-agent workflows
• Add API (FastAPI) for deployment
• Add UI (optional)

---

🧳 Portfolio Value

This project demonstrates:

• Real-world AI agent design
• Reliability-focused architecture
• Production-ready structure

---

👤 Author

Built by Diego Martínez Beltrami
AI Engineer (Computer Vision / Edge AI → Agentic AI transition)