A technical overview of the platform, its model architecture and the mechanisms that power our market analysis — without giving away the recipe.
KANNOT-1M is our proprietary detection model. A gradient-boosted ensemble of 5,034 decision trees trained on 224K+ data points and 79 features to classify market conditions into buy, hold or sell signals. Dedicated tag models identify regimes — bull, bear, stagnant — so every signal comes with context.
When the model flags a point of interest, TRADEMEM checks what was happening across a predefined set of reference assets at the same moment. Coverage verification runs across multiple timeframes to build a multi-dimensional view of correlation around each event.
Edges are predefined reference assets — stocks, ETFs, crypto, currencies — tracked alongside the primary analysis. The system maintains synchronized candle data across multiple timeframes, allowing the model to go beyond single-asset recognition and provide cross-market context.
TRADEMEM distills complex model output — probabilities, regime tags, coverage ratios — into clear, actionable information. Buy, hold or sell signals with confidence levels and correlation context, designed to reduce cognitive load.
The model identifies the nearest historical points to the current setup and looks at what happened next. Weighted percentile fusion merges the resulting price paths into a robust forward curve, filtering out outliers.