How Product-Fit Prediction Works Over Time

Prediction in TOQidex is not based on generic recommendations or population data. It is built entirely from your personal product experiences, making each prediction unique to you.

The prediction pipeline begins with your product logs. Each time you try a cannabis product and rate your experience, TOQidex records your 8-dimension scores alongside the product's full chemical profile: 13 cannabinoid concentrations and 34 terpene measurements.

After you have logged at least 10 products, TOQidex initiates model training. Three machine learning architectures are trained in parallel: a neural network (TensorFlow.js, 4-layer dense regression), a support vector machine (scikit-learn SVR with RBF kernel), and a gradient boosting regressor (scikit-learn).

Each model is evaluated on accuracy metrics including R-squared, mean squared error, and mean absolute error. The best-performing model is selected automatically. New model versions only replace existing ones if the accuracy improves.

When you encounter a product you have not tried, TOQidex feeds its chemical profile into your personal model to generate predicted scores across all 8 recreational dimensions and your tracked medical condition symptoms.

The more products you log, the better the predictions become. This is because the model has more data points to learn the specific cannabinoid and terpene patterns that correlate with your individual responses.

Predictions are presented as likely outcomes, not guaranteed results. Cannabis response is influenced by many factors beyond chemistry, including tolerance, consumption method, environment, and individual biology. TOQidex helps you build a structured evidence base for better decision-making, not a certainty engine.