INTER

LOCK

By Alpha31

Hana Benko, Laasya Aki, Dhiren Narne, Adithya Ravichandran

STEELHACKS XIII

2026

The planning problem

• Street changes are expensive, disruptive, and difficult to test before deployment.�
• Cities often rely on historical crash records, traffic counts, and static models.�
• Those approaches are poorly suited to evaluating rare near-miss interactions, temporary construction, weather conditions, and emerging AV traffic before a change is built.

​

STEELHACKS XIII

​

2026

• Lane or intersection closures during construction
• Signal timing and road redesigns
• Heavy rain, snow, or traffic surges

​

• Pedestrian safety and wait times
• Mixed human + autonomous traffic
• Before/after tradeoffs in safety and throughput

What they need plan for

What planners actually need to test

• Lane or intersection closures during construction
• Signal timing and road redesigns
• Heavy rain, snow, or traffic surges

​

Interlock is a scenario simulator for real Pittsburgh streets.

Real Pittsburgh Data

​

SUMO & Monte Carlo

​

Interactive 3D

​

AI Voice

​

• Pittsburgh traffic, crash, road, and weather data
• Stored in Tiger Data for simulation + analysis

• SUMO models vehicles, pedestrians, and signals
• Monte Carlo compares baseline vs. changed scenarios

• Three.js visualizes real intersections and simulation replays
• Users test construction, infrastructure, and AV scenarios

• Gemini explains results and scenario tradeoffs
• ElevenLabs enables voice interaction

Tech Stack

Python & SUMO Engine

Executes microscopic traffic flow modeling and Monte Carlo simulations to derive speed, delay, and near-miss metrics.

​

Data

Physics Engine

System Architecture

Analytics + Voice

• Ingest real Pittsburgh crash, traffic-count, road-design, and weather data
• Tiger Data / PostgreSQL stores normalized real-world data and high-frequency simulation results
• Local data calibrates traffic demand and baseline conditions

• Python + SUMO/TraCI simulate individual vehicles, pedestrians, signals, and road changes
• Monte Carlo trials compare baseline vs. modified scenarios
• Measures speed, delay, throughput, pedestrian wait, and Time-to-Collision conflict metrics

​

STEELHACKS XIII

2026

• FastAPI connects the interface to the engine
• ElevenLabs provides the voice interface and reads scenario/street information aloud
• Gemini interprets natural-language requests and translates them into model/scenario changes (only say this if this integration is actually working)

Data

Data

Demo!

STEELHACKS XIII

2026

Limitations & Next Steps

• Currently focused on a small number of Pittsburgh intersections
• Some traffic conditions are approximated where local data is limited
• Near-miss metrics estimate risk, not actual future crashes
• Autonomous vehicles are represented with simplified driving behavior
• Future work: connect many intersections into larger city-wide simulations

STEELHACKS XIII

2026

Hana Benko

Dhiren Narne

Adithya Ravichandran

Laasya Aki

Thank you!

hbenko@andrew.cmu.edu

dsnarne@andrew.cmu.edu

adithyar@andrew.cmu.edu

laki@andrew.cmu.edu