AI Traffic Control (AITC)

Inspiration

Air traffic control is one of the most complex and high-stakes professions in the world. Controllers must monitor dozens of aircraft simultaneously, interpret continuous streams of real-time data, and make split-second decisions using precise, standardized communication.

Instead of navigating multiple radar displays, flight strips, and decision-support tools, what if controllers could ask direct questions like “Which arrivals will conflict with runway 28 in the next 10 minutes?” or “Show me all aircraft descending through 10,000 feet in this sector” and get instant, context-aware answers?

This led to AI Traffic Control (AITC), an intelligent assistant that lets ATC controllers interact with real-time aviation data by asking questions like “Where is flight UAL123?” or “Show arrivals at JFK.” Instead of navigating complex dashboards, memorizing API parameters, and being overloaded with immense information, ATC controllers get instant answers with visual map overlays.

Our goal is to explore how AI agents can act as intelligent copilots for ATC that continuously monitoring live flight data, understanding controller intent, and surfacing relevant information at the moment it’s needed, without disrupting existing workflows.

What it does

AITC is a cross-platform desktop application that combines AI reasoning with live global flight tracking.

Natural language flight queries

Find flight DLH456
Show flights over California
What are the arrivals at Frankfurt airport?
Get the trajectory for AAL100

Real-time flight tracking (OpenSky Network API)

Live position, altitude, velocity, vertical rate, and heading
Callsign and ICAO24 identifier
Origin country and squawk code

Interactive map visualization

Leaflet-based world map with aircraft markers
Airplane icons rotate based on heading
Click-to-select aircraft for detailed information
Historical trajectory overlays

AI-powered intelligence

LangGraph-based agentic architecture
Automatic tool selection based on query intent
Context-aware responses with formatted flight data
Supports both cloud (Gemini) and local (Ollama / Llama 3.2) LLMs

Airport operations

Recent arrivals and departures at any airport
Automatic ICAO airport code resolution (JFK → KJFK)
Historical flight data up to 48 hours

How we built it

AITC is built as an Electron desktop application with a React frontend and a Python-based AI backend.

Frontend (React + TypeScript)

Main dashboard layout
Leaflet map for real-time aircraft visualization
Chat interface and flight detail panels
Natural language input component

AI backend (Python + LangGraph)

Seven specialized tools:
- search_flight_by_callsign
- get_current_flight_state
- get_flights_in_area
- get_aircraft_flight_history
- get_airport_arrivals
- get_airport_departures
- get_aircraft_trajectory
ReAct-style loop: interpret intent, select tool, execute, and synthesize results

Frontend–backend integration

Electron spawns a Python subprocess for each query
Natural language input is sent to the LangGraph agent
Formatted results update the UI and map in real time

Privacy-first design

Google Gemini for cloud inference
Ollama (Llama 3.2) for fully local inference

Challenges we ran into

API rate limiting required throttling, request queuing, caching, and fallback data
Real-time map updates required careful synchronization and animation handling
Natural language variability required regex-based intent parsing and airport aliases
Reliable LLM tool calling required careful prompt and system message design