Model Architecture
Interactive visualization of our 8 production ML models powering F1 predictions
Beyond F1 Team Tools:Custom NBT-TLF architecture, temporal embeddings, and multi-head attention - capabilities not available in standard F1 analytics systems.
Select Model to Explore
Neural Bradley-Terry with Temporal Latent Factors
Our flagship custom architecture - a neural pairwise ranking model with learned temporal embeddings. Not available in any ML library.
💡 Innovation
🚀 First-of-its-kind neural Bradley-Terry implementation with sinusoidal positional encoding for season progression.
0.950
AUC-ROC Score
45 min
Training
12ms
Inference
Custom Neural NetworkComplexity: High
Neural Network Architecture(Click layers to explore)
15
Input
32
Driver
32
Constructor
32
Track
16
Temporal
64
Multi-Head
128
Hidden
64
Hidden
32
Hidden
1
Output
What Makes Our Models Unique
🧠
Custom NBT-TLF
Neural pairwise ranker not available in any ML library
⏱️
Temporal Embeddings
Captures season progression and form trajectory
🎯
Multi-Head Attention
4-head attention for driver-track interactions
📊
8-Model Ensemble
Diverse models for robust predictions