🚗 Merge
Ride together. Park smarter.
Inspiration
Every morning we see the same inefficiency:
- Cars driving with empty seats
- Drivers circling blocks for parking
- Private driveways sitting unused
Carpooling apps exist. Parking marketplaces exist. But they operate separately.
We asked ourselves:
Why are movement and parking treated as two different problems?
A commute has two phases:
- Getting there
- Storing the vehicle
Merge was inspired by the idea that these should be optimized together.
What It Does
Merge is a neighborhood mobility intelligence system that coordinates:
- 🚘 Carpool matching
- 🅿️ Private parking allocation
- 📊 Demand prediction
- 💵 Dynamic pricing
- 🤖 Self-improving strategy updates
Instead of solving isolated problems, Merge treats the city as one coordinated system.
We define overall system efficiency as:
$$ E = \frac{R + U}{D + C} $$
Where:
- (R) = Ride completion rate
- (U) = Parking utilization
- (D) = Average delay
- (C) = Cancellation rate
Our goal is to maximize (E) over time.
How We Built It
Merge is designed as a multi-agent system.
Ride Matching Agent
Pairs drivers and riders based on:
- Route overlap
- Time flexibility
- Reliability score
Parking Allocation Agent
Matches vehicles with available private parking spots using:
- Walking tolerance
- Budget sensitivity
- Availability window
Reliability Agent
Tracks:
- On-time arrival percentage
- Cancellation frequency
- Ratings
This agent produces a dynamic trust score used across the system.
Demand Prediction Agent
Forecasts:
- Peak congestion
- Parking scarcity
- Zone-based demand spikes
Reflection Agent
At the end of each simulated day, the system analyzes:
- Failed rides
- Parking conflicts
- Delay clusters
- Underutilized zones
It generates structured insights like:
“Zone 3 morning rides fail due to underestimated traffic variability.”
Strategy Update Agent
Using reflection outputs, this agent updates system parameters:
- Matching weights
- Trust thresholds
- Surge pricing triggers
- Route buffer times
This is what makes Merge agentic and self-improving.
The Agent Loop
Each cycle:
- A simulated day runs
- Metrics are collected
- Reflection occurs
- Strategy updates are applied
- The next day runs with improved coordination
Over time, performance improves.
Example progression:
| Day | Ride Completion | Parking Utilization | Avg Delay |
|---|---|---|---|
| 1 | 62% | 55% | 14 min |
| 3 | 78% | 70% | 9 min |
| 5 | 89% | 83% | 5 min |
Technical Design
We structured the optimization objective as:
$$ C = w_1R + w_2U - w_3D - w_4X $$
Where:
- (R) = Ride success
- (U) = Parking utilization
- (D) = Delay
- (X) = Cancellation rate
- (w_i) = Adaptive weights
The Strategy Agent updates the weights (w_i) after each cycle to improve coordination.
What We Learned
- Coordination is harder than simple matching.
- Reliability has a cascading effect on system performance.
- Reflection is critical for measurable improvement.
- Urban mobility is fundamentally a systems problem.
Without feedback loops, intelligent systems stagnate.
Challenges We Faced
Simulating Real Urban Behavior
We had to create synthetic commute zones, traffic variability, and user behavior patterns.
Balancing Complexity
We initially over-designed the agent system. We refined it to maintain clarity while preserving intelligence.
Making Improvement Measurable
We needed clear metrics to demonstrate system evolution across iterations.
Why Merge Matters
Merge demonstrates that infrastructure can learn.
If cities coordinate movement and storage together:
- Traffic decreases
- Emissions drop
- Idle space becomes productive
- Commutes become less stressful
Merge is a prototype of adaptive urban coordination.
Built With
- amazon-bedrock
- claude
- expo.io
- google-cloud
- kiro
- openai
- react-native
Log in or sign up for Devpost to join the conversation.