Atlas

Inspiration

Airline disruptions get messy fast. A single weather event at a major hub can ripple across the network, essentially delaying passengers, stressing nearby airports, and forcing operators to make hard trade-offs in real time. Most tools either oversimplify the problem or dump raw technical data on users who don’t have time to interpret it.

We built Atlas to feel more like how airport and airline operations teams actually think: understand the disruption, explore recovery options, and clearly see the impact of each choice—without needing to know anything about the underlying compute or algorithms.

What it does

Atlas is an airline disruption recovery and route optimization simulator.

Given a disruption (like severe weather at one or more airports), Atlas:

Pulls in a baseline flight schedule
Models disruption severity and airport capacity limits
Explores hundreds of recovery strategies in parallel
Recommends the best recovery plans for different goals:
- Fastest recovery
- Passenger-first
- Lowest cost
- Fewest cancellations
- Crew-safe
Shows flight-level impacts with clear before-and-after routes
Visualizes airport capacity changes, so traffic is redistributed realistically instead of overwhelming a single hub

Most flights end up delayed, some are rerouted through nearby hubs when it’s faster than waiting, and cancellations are treated as a last resort—just like in real operations.

How we built it

Atlas uses a parallel simulation setup:

Backend (Python)
- Each disruption scenario is a job
- Each job runs hundreds of greedy, heuristic-based simulations in parallel
- Every simulation produces one feasible recovery plan
- Plans are ranked after simulation using user-defined priorities
- Airport capacity, delay propagation, disruption severity, and rerouting logic are all explicitly modeled
Frontend (React + TypeScript)
- A guided, narrative-driven UI that starts with the disruption
- Clear recovery plan comparisons without technical jargon
- Flight-level views with filters for affected airports
- Airport capacity dashboards showing baseline vs recovered utilization
GenAI-assisted compute planning
- GenAI translates user intent (speed vs cost vs accuracy, natural language preferences) into a compute plan
- The actual optimization stays deterministic and explainable

Where real-world data isn’t available, we use dummy but plausible data so everything stays consistent and demo-ready.

Challenges we ran into

Balancing realism with something we could actually build in a hackathon
Making sure reroutes and delays only happen when there’s a real reason
Keeping baseline and recovery plans consistent so capacity numbers make sense
Designing a UI that explains what’s happening without overwhelming users

Accomplishments that we're proud of

Reroutes that actually make sense—only when they beat delaying, and always through nearby hubs
Capacity-aware recovery plans that avoid unrealistic “magic” solutions
A UI that’s understandable even if you’ve never seen an optimization tool before

- The simulation is purely algorithmic and heuristic so... it's fast

What we learned

Optimization only matters if people can understand and trust the results
Constraints and guardrails matter more than clever tricks
GenAI works best as a planner and coordinator, not a replacement for domain logic
The strongest hackathon projects feel like real products, not research prototypes