✈️ Flight Intelligence Automation Agent

A local AI-powered automation system that transforms flight sales data into actionable insights — built on top of your Flight Sales Data Pipeline project using DuckDB, Airflow, Streamlit, and open-source LLMs orchestrated via the Model Context Protocol (MCP).

🧭 Overview

This project extends the previous Airflow ETL pipeline into a full AI automation system.
It connects flight sales data (simulated via Mock API) with a local Hugging Face LLM to:

Generate daily business insights from flight data.
Automate summaries and anomaly detection.
Store and visualize insights in a Streamlit dashboard.
Run fully locally, using the MCP framework for modular automation.

⚙️ Tech Stack

| Layer | Tool | Purpose | |-------|------|----------| | 🐍 Programming | Python | Core scripting language | | 🪶 Storage | DuckDB | Analytical database | | 🧱 Orchestration | Apache Airflow | ETL and automation | | 🧠 AI Model | Hugging Face (Phi-2 / Mistral / Llama) | Insight generation | | 🔗 Protocol | MCP (Model Context Protocol) | Tool orchestration | | 📊 Visualization | Streamlit | Interactive dashboards | | 🪄 Automation | Watchdog / Cron | Local triggers for automation | | 🐳 Containerization | Docker | Optional Airflow deployment |

📂 Folder Structure

flight_intelligence_agent/
├── db/                             
│   ├── raw_data.duckdb
│   ├── db_connection.py
│   ├── init_schema.py
│   ├── fetch_and_insert.py
│   └── transformations.py
│
├── airflow/                        
│   ├── dags/
│   │   └── etl_pipeline.py
│   └── docker-compose.yaml
│
├── agents/
│   ├── insights_agent.py
│   └── scheduler.py
│
├── mcp_layer/
│   ├── client.py
│   └── servers/
│       ├── duckdb_server.py
│       ├── filesystem_server.py
│       └── ai_server.py
│
├── dashboard/
│   └── app.py
│
├── reports/
│   ├── daily_summary.txt
│   └── anomalies.json
│
├── config/
│   ├── config.json
│   └── .env
│
├── requirements.txt                
├── .gitignore
└── README.md

🧩 Architecture

Architecture Diagram

🔍 How It Works

🛫 Data Ingestion

fetch_and_insert.py fetches flight data from the Mock API.
Airflow DAG (etl_pipeline.py) automates the daily ETL process (Extract → Transform → Load).
Data is stored in DuckDB under the schema raw.bookings.

🔄 Data Transformation

transformations.py computes KPIs such as total revenue, average ticket price, and unique passenger count.
Processed data is stored in the processed schema for analysis.

🧠 AI Insights Layer

The MCP AI Agent (insights_agent.py) connects to DuckDB and retrieves KPIs.
It feeds the data into an open-source Hugging Face model (e.g. Phi-2 or Mistral).
The agent generates natural-language summaries and anomaly alerts, saved in /reports/.

⚙️ Automation via MCP

The MCP Client orchestrates communication between multiple local servers:

| Server | Purpose | |---------|----------| | 🪶 duckdb_server.py | Exposes SQL queries and KPI data. | | 📁 filesystem_server.py | Manages local report generation and file monitoring. | | 🧠 ai_server.py | Wraps a Hugging Face LLM for text generation and inference. |

➡️ The AI Agent acts as the central orchestrator, dynamically calling these servers through the Model Context Protocol.

📊 Visualization

The Streamlit dashboard reads processed data from DuckDB and insights from /reports/.
Displays real-time KPIs, sales trends, and AI-generated summaries.

🧠 Example Output

Example AI Summary:

Revenue increased 8% week-over-week, primarily driven by Business Class bookings from Paris → New York.
However, Economy fares saw a 5% drop in occupancy rate due to seasonal effects.

🚀 Quickstart

1️⃣ Setup Environment

git clone https://github.com/<your-username>/flight_intelligence_agent.git
cd flight_intelligence_agent
pip install -r requirements.txt

2️⃣ Run ETL Pipeline

python db/init_schema.py
python db/fetch_and_insert.py

Or start with Airflow:

docker-compose up

3️⃣ Run AI Insight Agent

python agents/insights_agent.py

4️⃣ Launch Dashboard

streamlit run dashboard/app.py

MCP Servers