Back to Dashboard

The Science of Prediction

Traffic Prediction with Contextual Intelligence

Standard GPS apps just look at "current speed". We look deeper. Our system analyzes dozens of variables to find historical patterns that match the exact DNA of the current moment.

The Context Engine

Cyclical Time

We don't just see linear time. Our model understands the cyclical nature of rush hours, knowing that 11 PM flows naturally into Midnight.

Dayflow Analysis

Fridays aren't just 'Weekdays'. Our system distinguishes the unique traffic signature of a Friday afternoon vs. a Monday morning.

Seasonal Trends

Summer tourist traffic behaves differently than winter commuter traffic. We adapt to the month and season automatically.

Event Urgency

Is the game starting in 30 minutes? Or did it just end? We track the 'Pulse' of the city to predict sudden surges.

Crowd Typing

Not all crowds are the same. We differentiate between the steady flow of a Marathon vs. the sudden burst of a Concert exit.

Impact Magnitude

We assess the size of every event—from local gatherings to massive stadium shows—to weight their influence on the border.

Weather Resistance

Rain slows traffic. Snow stops it. Our model finds historical days with similar weather profiles to adjust the forecast.

Origin Tracking

Where is the crowd coming from? We analyze if an event pulls traffic North (towards Canada) or South (towards USA).

Holiday Flow

Canadian holidays push traffic South. US holidays push it North. We account for these massive directional shifts.

The Twin-Track Strategy: Planner vs. Nowcast

Strategic Layer

Context Vectors (Planning)

Used for the 7-Day Forecast. This track searches for historical matches based on "Static Context"—long-range factors like holidays, stadium event schedules, and seasonal weather patterns.

Goal: Accuracy at Distance

"What is the most likely shape of traffic next Tuesday?"

Tactical Layer

Dynamics Vectors (Nowcasting)

Powering Live Updates. This track ignores the calendar and matches current "Traffic Physics"—momentum, acceleration (rate of change), and spatial variations between neighboring ports.

Goal: Real-time Resilience

"Something is happening right now. Where have we seen this physics before?"

From Data to Action: Applied Generative AI

Beyond Numbers:
An AI That Thinks Like a Local

Raw data charts are hard to read. We use DSPy (Declarative Self-Improving Python) to transform complex vector statistics into clear, actionable advice.

  • Self-Correcting LogicUses a "Generate & Filter" architecture. If the AI drafts a report that contradicts the data (e.g., calling a 90-minute wait "smooth"), it catches its own mistake and rewrites it automatically.
  • Smart Timing RecommendationsInstead of static graphs, it finds specific "Green Windows"—identifying exactly when you should leave to shave 30+ minutes off your trip.
INPUT: Context JSON
data: { peak_wait: 90, event: "Concert" }
GENERATION 1 (Rejected)
"Traffic looks good! Head out now."
GENERATION 2 (Approved)
"High traffic alert! Wait until 8 PM to avoid post-concert gridlock."

Ex
Real-World Scenarios

Time Match
Weather Match
Event Match
Prediction Confidence
High

Scenario: "The Rainy Friday Concert"

Imagine a major concert ending on a rainy Friday night. Simple averages would just look at "Friday Night Traffic".

Our system searches specifically for past rainy Fridays with major events. It knows that rain + exiting crowds creates a unique congestion pattern that normal Fridays don't have.

Scenario: "Holiday Directionality"

Not all holidays are the same. US Thanksgiving creates a massive surge Southbound, but Canada Day surges Northbound.

Our engine applies a Directional Bias to every holiday, ensuring we compare apples to apples. If it's Canada Day, we ignore regular Monday traffic and look for similar Northbound-heavy historical days.

Upcoming Day
Canada Day 🇨🇦
MATCHES
Historical Match
Past Canada Day 2024