Predict your peak

Powered by neural networks and machine learning, FitPredict transforms your biometrics into real-time precision nutrition, hydration, and training prescriptions.

Start your analysis

Personalized Precision Plan

Set your training goal and adjust your biometrics on the right-hand panel, then run the analysis to instantly generate your precision nutrition and recovery plan.

Body & Routine Metrics

Adjust parameters for your target plan
Body Biometrics
Physique
Weight kg
74
Height cm
178
Age yrs
28
Training & Routine
Schedule
Sleep hrs
7.5
Workouts /wk
4
Intensity RPE
7

Neural Network Engine

How our multi-task deep neural network transforms continuous physiological telemetry into sub-second adaptive training prescriptions.

01

Multi-Modal Feature Tensors & Vector Embeddings

Biometric telemetry undergoes z-score normalization to map high-dimensional non-linear physiological relationships.

Tensor Normalization Latent Space Mapping ACSM Vector Bounds
02

Deep Multi-Task Neural Backbone (MLP-ResNet)

Deep feedforward layers with residual connections compute shared representations for simultaneous multi-target predictions.

Residual Skip Connections GELU Activations Multi-Target Regression
03

Sub-12ms Client ONNX Runtime & Adaptive Loss Tuning

Executes in-browser in under 12ms via optimized ONNX, calibrated on 10,000+ clinical datasets for guaranteed safety.

< 12ms Inference Latency Huber Loss Calibration Continuous Safety Bounds

Calibrated Athlete Archetypes

Hover across the cards to explore athletic disciplines calibrated with clinical exercise science. Click any preset to instantly load its biometrics into the predictor.

Swipe or drag the cards horizontally to spin the turntable ring

Model Accuracy & Performance

Empirically evaluated using 10-fold cross-validation on 10,000+ physiological datasets against clinical exercise trials.

0.0%

Test Accuracy

Cross-validated concordance across multi-target physiological prediction outcomes

0.00

Mean Absolute Error

Normalized loss across multi-target continuous regression vectors

0.00

Score (R²)

Variance accounted for by neural latent embeddings against clinical trials

< 0ms

Inference Latency

Sub-millisecond real-time execution compiled via optimized ONNX WebAssembly

How FitPredict Empowers Your Performance

Our central physiological AI engine autonomously coordinates synchronized performance domains in real time, turning raw telemetry into dynamic training adaptations.

AI

FitPredict AI Core

Neural Orchestrator

Adaptive Nutrition

Sub-gram protein pacing and metabolic fuel custom-tailored to body mass and training load.

Dynamic Recalibration

Sleep shifts and workout RPE immediately adjust your training prescription in sub-seconds.

Hydration Dynamics

Osmotic fluid replenishment and electrolyte pacing computed from real-time sweat rate.

Test Your Personal Targets Now

Open Source Neural Architecture

We believe athletic science should be transparent and reproducible. Inspect the complete multi-task neural network pipeline, benchmark datasets, and client runtime directly on GitHub.

MIT License
git clone https://github.com/fitpredict/neural-engine.git

fitpredict / neural-engine

Public Repository
v2.4.0
View on GitHub