Back to Home

The Science Behind Parlay Buddy

We don't predict games. We predict you.

11 Iterations per parlay
Pattern combinations
0 Team stats required
AI

Why Traditional Models Fail

Stats Are Already Priced In

Team efficiency, passing yards, injury reports—bookmakers already use this data. Your model just confirms what Vegas already knows.

Model says: "Favorite wins" → Odds: -350 No edge gained

Upsets Are Non-Statistical

The factors that cause upsets—motivation, weather momentum, coaching decisions—aren't captured in spreadsheets.

Data shows: Team A 78% win rate Team B wins anyway

Our Approach

Instead of predicting games, we predict YOUR decision patterns—when you take risks, when you play safe, and what that means.

Pattern detected: Risk escalation in legs 7-11 Actionable insight

How Parlay Buddy Works

Three steps to pattern-aware predictions

1
F F U
2
F U F
3
U U F
...
11
U F U
01

Fill 11 Iterations

Pick the same games 11 different ways. Each iteration captures a different risk tolerance—conservative to aggressive.

  • Same matchups, different combinations
  • Captures your decision variance
  • Reveals hidden biases
02

Transformer Analysis

Our neural network processes all 11 iterations simultaneously using self-attention to find relationships you can't see.

  • Every leg compared to every other leg
  • Pattern recognition across sequences
  • Psychological stress point detection
Self-Attention Weights
Lakers vs Celtics 87%
💡 You pick underdogs 73% when odds exceed +200
Chiefs vs Bills 64%
⚠️ High variance in late iteration picks
03

Pattern-Aware Predictions

Get personalized predictions based on YOUR betting patterns, not generic statistics.

  • Confidence scores per matchup
  • Upset tendency insights
  • Historical pattern alignment

The Technical Architecture

Transformer Encoder

Unlike RNNs that process sequentially and "forget" earlier inputs, our transformer sees all 11 iterations at once.

RNN/LSTM
Sequential, loses context
Transformer
Parallel, full context

Self-Attention Mechanism

Each betting leg "attends" to every other leg, learning which combinations matter most for YOUR patterns.

Leg 1
Leg 2
Leg 3

Behavioral Feature Engineering

We extract 20+ features from your picks that bookmakers don't track:

  • Position Where upsets appear in sequence
  • Variance How much you change across iterations
  • Clustering Which picks always go together
  • Momentum Risk escalation patterns

The Key Insight

"We don't predict who wins the game. We predict how YOU will pick—and whether that pattern historically leads to success."

Traditional models ask: "Will Team A beat Team B?"
We ask: "When you pick Team A as an underdog in iteration 7, after picking them as favorite in iterations 1-3, what does that pattern mean?"

What We Analyze

01

Iteration Position

Which iteration contains each pick. Early iterations often reflect "gut" picks, later ones show reconsideration.

02

Upset Distribution

Where you place underdogs across your 11 iterations. Clustering reveals confidence levels.

03

Flip Frequency

How often you change a pick between iterations. High flip rate on a leg signals uncertainty.

04

Risk Gradient

Whether your picks get riskier or safer as iterations progress. Most bettors show escalation.

05

Combination Locking

Which picks always appear together. These "locked" combinations reveal your core beliefs.

06

Odds Sensitivity

How line movements affect your choices. Some bettors chase value, others avoid it.

What We Don't Use

These traditional data points are intentionally excluded because they provide no edge:

  • Team Statistics Already priced into odds by bookmakers
  • Player Performance Data Public information with no hidden value
  • Historical Head-to-Head Past matchups don't predict future upsets
  • Injury Reports Instantly reflected in line movements

Traditional Approach

Team Stats
ML Model
Predicts Favorite
Same conclusion as Vegas
VS

Parlay Buddy

Your Patterns
Transformer
Predicts YOU
Unique behavioral edge

Future Enhancements

What we're building next

Deep Q-Learning Agent

An AI that learns your risk/reward preferences and suggests optimal picks using reinforcement learning.

Q3 2026

Live Game Analysis

Real-time pattern matching during games. See how your in-game instincts compare to pre-game picks.

Q4 2026

Custom Model Training

Train specialized models for specific leagues, bet types, or time periods.

2027

Your Data, Your Control

We only store your picks—no personal betting amounts, no sportsbook connections, no data sales.

No Gambling Here

Parlay Buddy is an analysis tool. You cannot place bets through our platform.

Honest Accuracy

We display YOUR model's win rate—no cherry-picked stats. Results vary by sport and season.

Ready to Discover Your Patterns?

Stop guessing. Start understanding how you bet.

Pattern-Aware AI No Team Stats Needed Built with PyTorch