Inspiration
Living in Texas, we've experienced firsthand how severe winter storms can ground flights and strand thousands of passengers at DFW. Just this January, weather disruptions caused chaos across the entire hub network. We wanted to build a tool that could help airport operations centers respond instantly to weather emergencies—turning complex optimization into a single button press.
What it does
TAMUFlights is a real-time flight disruption management dashboard that helps aviation operators reroute flights around severe weather.
Key Features:
Live Flight Tracking - Visualizes 20-50 aircraft in the DFW region using real FlightAware data Real-Time Weather Radar - Shows actual precipitation, clouds, and storms on an interactive map AI Route Optimization - Uses Dijkstra's algorithm to automatically find safe alternate routes around weather GPU Acceleration - Detects NVIDIA GPUs and switches to HPC Mode automatically (falls back to CPU on Macs) Smart Priorities - Emergency flights (medical evac, organ transport) get the best routes AI Explanations - Google Gemini explains every decision in plain English "God Mode" - Inject simulated storms to test system responses during demos How we built it Frontend: React + Vite, Mapbox GL JS for maps, Tailwind CSS for the glassmorphic UI
Backend: Python FastAPI, SQLite database, NetworkX for graph algorithms
Data Structure: We model the airspace as a directed graph where airports are nodes and flight routes are edges. Each edge has a weight based on distance and weather severity.
Algorithm:
Build a graph of all airports When storms appear, increase edge weights on affected routes (severity 10 storm = 1024x penalty) Run Dijkstra's shortest path algorithm for each flight Emergency flights get priority with lower cost weights Send results to Gemini for AI-generated explanations GPU Detection: The system runs nvidia-smi to detect NVIDIA GPUs (HPC Mode) or checks for Apple Silicon (Standard Mode), then displays the detected hardware in the UI.
Challenges we ran into
API Quota Limits - FlightAware only allows ~100 calls/month on the free tier. We implemented a cache-first architecture with manual refresh buttons to avoid burning through our quota.
Route Visualization - Routes initially showed as straight lines even with storms active. We fixed this by checking multiple points along each flight path (not just the midpoint) to properly detect storm collisions and curve routes around them.
Cross-Platform GPU Support - Our team had Windows laptops with RTX 5070, MacBooks with M3 Pro, and Linux VMs. We built a multi-stage detection system that works everywhere and shows what hardware is being used.
Weather Layer Performance - Radar overlays were hiding flight markers. We added opacity controls and a Weather ON/OFF toggle button.
Accomplishments that we're proud of
Built a full production-grade system with real APIs in 24 hours
Automatic GPU detection and acceleration—no manual configuration needed
Real live flight and weather data, not mock data
AI-generated explanations make complex decisions transparent
Glassmorphic UI that looks like a professional air traffic control system
Applied textbook graph theory (Dijkstra's algorithm) to solve a real-world problem
What we learned
How to model real-world problems (airport networks) as graphs and apply optimization algorithms GPU acceleration with NetworkX and RAPIDS cuGraph Working with geospatial APIs and tile-based map overlays The importance of API rate limiting and quota-safe architecture design Flight operations prioritization—emergency medical flights must get optimal routes even if it costs 100x more fuel Cross-platform development and building systems that work on any hardware What's next for TAMUFlights
Short-term:
Add crew scheduling constraints and gate availability Show passenger impact scores for each reroute Expand to other hub networks (ORD, ATL, LAX) Long-term:
Machine learning to predict storm paths 2-4 hours ahead Mobile app so passengers can see AI explanations for delays Partner with airlines like American Airlines to deploy in real operations centers Scale to handle 10,000+ flights using cloud GPU clusters
Log in or sign up for Devpost to join the conversation.