BTC/EUR --
ETH/EUR --
Meridian Labs

4 architectures.
One consensus.

Four independent AI architectures. One consensus-driven decision. Live on Kraken and OKX.

276 Models · 4 Architectures · 37 EUR Pairs · Live
Signal Pipeline

From Data to Execution

Every trade passes through a multi-stage pipeline. Raw market data is transformed into high-confidence signals, validated by an independent LLM, before capital is deployed.

44 FEATURES 4-MODEL ENSEMBLE CONSENSUS โ†’ EXECUTE
01
Data Ingest
OHLCV candles for 37 EUR trading pairs. Order book depth, funding rates, and market microstructure signals collected continuously.
02
Feature Engineering
44 engineered features extracted from raw data โ€” technical indicators, volatility measures, volume profiles, and cross-pair correlations.
03
4-Model Inference
276 models across 4 architectures (LSTM, Transformer, XGBoost, LightGBM) generate independent directional signals per pair.
04
Ensemble Voting
Signals aggregated by majority vote. 3/4 or 4/4 model agreement required to generate a tradeable signal. 2/4 or less = no trade.
05
Caelum-8B Validation
Fine-tuned local LLM (caelum-8b) provides independent second-opinion validation on every consensus signal before execution.
06
Execution
Orders placed on Kraken and OKX via API with risk filters, slippage control, and real-time position tracking every 3 minutes.
Architecture

Multi-Model Ensemble

Four fundamentally different model architectures. Each sees the market from a different angle. When they agree, the signal carries conviction no single model can achieve.

🧠

LSTM Neural Networks

72 LSTM models trained on 6 months of historical data. Specializes in sequential price patterns and short-term momentum detection.

SEQUENTIAL
💡

Transformer Encoder

Attention-based models that evaluate all data points simultaneously. Captures long-range dependencies and complex cross-feature relationships that sequential models miss.

ATTENTION
🌳

XGBoost

Gradient boosted decision trees. Excels at structured tabular data and engineered features. Often outperforms deep learning on financial data.

GRADIENT BOOST

LightGBM

High-performance gradient boosting optimized for speed and large feature sets. Provides a complementary perspective to XGBoost with different splitting strategies.

FAST BOOST
Consensus

Ensemble Voting

When four fundamentally different architectures independently reach the same conclusion, that signal carries conviction no single model can achieve.

4/4 Agree

Highest conviction trade. All four architectures independently reached the same conclusion.

HIGHEST CONVICTION

3/4 Agree

High conviction trade. Strong majority consensus across fundamentally different model types.

HIGH CONVICTION

2/4 or Less

No trade. Insufficient consensus. Capital protected. The system only acts when the evidence is overwhelming.

NO TRADE
Live Performance

Engine Status

Spot Engine LONG
Venue: Kraken
Loading...
Margin Engine SHORT
Venue: Kraken
Loading...
Copy Trading LEADERBOARD
Venue: OKX
Loading...

Recent Closed Trades

Loading trade history...
Risk Architecture

Capital Protection First

Hard-coded risk limits at every level. No override. No exceptions. The system protects capital before it seeks returns.

60%
Capital Reserve (Untouchable)
2%
Position Size Per Trade
-5%
Stop Loss Per Trade
-7%
Rolling 24h Drawdown Breaker
3 min
Health Check Interval
37
EUR Pairs Monitored
Infrastructure

Built for Reliability

Production-grade infrastructure with automated monitoring, self-healing watchdogs, and zero-downtime nightly retraining.

Python PyTorch LSTM Transformer Encoder XGBoost LightGBM Ensemble Voting Kraken API OKX API Caelum-8B (Ollama) RTX 4070 SUPER Nightly Retrain Regression Gate Watchdog Telegram Alerts Cloudflare