SkyLink Navigator - Agent for Air-Traffic Coordination

Inspiration

The inspiration came from the high-stakes nature of air traffic control, where even small delays or miscommunications can cascade into major risks. Major airports can handle over 1,200 flights a day, and globally, dozens of runway incursions or near misses each year are attributed to ATC–pilot miscommunication many of them preventable.

What it does

Our product is an AI-powered assistant for Air Traffic Controllers that:

• Listens to pilot requests (e.g., “Requesting to land”)
• Analyzes live flight schedules for delays, gate conflicts, or runway congestion
• Checks real-time geospatial data to detect off-course or deviated flight paths
• Evaluates current weather conditions and risk levels along the flight’s route
• Synthesizes a risk-aware response — recommending safe actions like landing clearance, holding, or rerouting
• Responds in natural language, like a smart tower assistant ready 24/7

How we built it

We built a multi-agent system using: Databricks Llama 4 as the core LLM (via databricks-genai) LangChain, LangGraph agents to wrap tools for schedule, geospatial, and weather checks LangGraph to create an LLM-driven workflow that loops through tool calls SQLite to simulate live flight, geo, and weather data. Conversation memory for continuous chat via ConversationBufferMemory Developed and run in Databricks notebooks, ready for Model Serving or Agent Playground use. Tech stack - - Unity Catalog, Langgraph, Model Serving, MLflow evaluate, Serverless, Vector Store, Genie

Challenges we ran into

Challenges included coordinating multiple tools (schedule, geo, weather), syncing simulated flight data across databases, and tuning the LLM to interpret pilot-style messages. Handling tool execution flow in LangGraph and deploying the agent for chat in Databricks required custom logic and prompt design. The result is a smart co-controller for safer airspace ops.

Accomplishments that we're proud of

• Built a fully working multi-agent ATC system in a short hackathon window
• Seamlessly integrated LLM reasoning with live tool data (schedule, geo, weather)
• Enabled the agent to handle natural pilot-style inputs and infer the right actions
• Achieved synchronized simulation across multiple databases
• Deployed the agent to Databricks Playground for live interactive chat
• Showed how AI can bring real-time decision support to a critical domain like aviation safety

What we learned

We learned how to build a multi-agent LLM system that combines real-time data, structured tools, and natural language understanding. We gained hands-on experience with LangGraph, Databricks LLMs, and deploying agents in production-like environments for complex decision-making.

What's next for SkyLink Navigator - Agent for Air-Traffic Coordination

Next, we aim to integrate real flight and weather APIs, expand to multi-airport coordination, and enhance decision-making with anomaly detection models. We also plan to add voice input for live ATC communication, build a real-time dashboard, and improve explainability and safety for deployment in critical airspace systems.

Built With

• databricks
• langchain
• langgraph
• llm