A production-grade intelligent trading analysis and execution platform that combines 400+ classical technical strategies with modern machine learning for automated signal generation, market regime detection, and risk-optimized trade execution.
The platform uses an ML ensemble pipeline (Random Forest, XGBoost, LightGBM) trained with time-series cross-validation on 50+ engineered features to predict price movements. A recommendation engine personalizes strategy selection based on market regime, while an NLP module extracts sentiment signals from market text. All components feed into a decision optimizer that computes Kelly-criterion position sizing.
Built for Forex, Crypto, Metals, and Indices via MetaTrader 5 with a FastAPI backend, real-time WebSocket updates, and a full-featured web frontend.
- ML Signal Generation — Ensemble of Random Forest, XGBoost, and LightGBM classifiers with
TimeSeriesSplitcross-validation and Optuna hyperparameter optimization - Market Regime Detection — Classifies current conditions (trending up/down, ranging, volatile) using ADX, ATR percentile, and linear regression slope analysis
- Intelligent Strategy Recommendation — Performance-weighted scoring with regime affinity, recency decay, and collaborative filtering from user trade history
- NLP Sentiment Analysis — Lexicon-based market sentiment extraction from news headlines with entity recognition (symbols, indicators, price levels)
- AI-Optimized Position Sizing — Fractional Kelly criterion with volatility adjustment, portfolio risk caps, and correlation-aware exposure management
- 400+ Technical Strategies — Rule-based strategies (RSI, MACD, Bollinger, Ichimoku, Smart Money, etc.) enhanced by ML confidence overlays
- Automated Trade Execution — Dual-server architecture with signal queuing, whitelist filtering, and adaptive risk controls
graph TD
A[Market Data — MT5] --> B[OHLCV Data Fetch]
B --> C[Technical Indicators<br/>50+ indicators]
C --> D[Feature Engineering<br/>50+ ML features]
D --> E[ML Prediction Engine]
E --> E1[Random Forest]
E --> E2[XGBoost]
E --> E3[LightGBM]
E1 --> F[Ensemble Aggregation]
E2 --> F
E3 --> F
C --> G[400+ Rule-Based Strategies]
G --> H[Strategy Orchestrator]
B --> I[Market Regime Detector]
I --> J[Recommendation Engine]
H --> J
F --> J
J --> K[Decision Optimizer<br/>Kelly Criterion + Risk Mgmt]
L[News / Text] --> M[NLP Sentiment Analyzer]
M --> K
K --> N[Signal Output<br/>BUY / SELL / HOLD]
N --> O[Trade Execution — MT5]
N --> P[Alert System<br/>Telegram / Email / Desktop]
N --> Q[Web Dashboard<br/>Real-time Updates]
# Clone the repository
git clone https://github.com/ARASH3280ARASH/ai-mvp-platform.git
cd ai-mvp-platform
# Create virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txtpython -m uvicorn backend.api.server:app --host 0.0.0.0 --port 8000from ai_core.pipeline import AIPipeline
from ai_core.config import AIConfig
# Initialize pipeline
pipeline = AIPipeline(config=AIConfig(), account_balance=10_000)
# Train on historical data
pipeline.train(df=ohlcv_data, indicators=indicator_dict, strategy_history=trades_df)
# Run real-time analysis
result = pipeline.analyze(
symbol="BTCUSD", timeframe="H1",
df=latest_data, indicators=indicators,
news_texts=["Bitcoin surges on ETF inflows"],
)
print(f"Signal: {result.ml_signal} | Confidence: {result.ml_confidence:.1%}")
print(f"Regime: {result.regime.regime} | Lot Size: {result.trade_decision.lot_size}")python bot_server.pyai-mvp-platform/
├── ai_core/ # AI / ML pipeline
│ ├── pipeline.py # Orchestrates all AI components
│ ├── predictive_model.py # Ensemble ML signal prediction
│ ├── recommendation_engine.py # Regime-aware strategy recommendation
│ ├── nlp_processor.py # Market sentiment analysis
│ ├── decision_optimizer.py # Kelly-criterion position sizing
│ ├── config.py # Model hyperparameters
│ └── utils.py # Feature engineering utilities
│
├── backend/ # FastAPI server
│ ├── api/ # 35+ REST endpoints + WebSocket
│ │ ├── server.py # Main application
│ │ ├── database.py # SQLite (WAL mode)
│ │ ├── alert_manager.py # Alert CRUD & notifications
│ │ ├── performance_monitor.py # Account metrics
│ │ ├── risk_engine.py # Risk calculations
│ │ └── ...
│ ├── strategies/ # 400+ trading strategies
│ │ ├── orchestrator.py # Strategy loader & dispatcher
│ │ ├── rsi_strategies.py # RSI variants
│ │ ├── macd_strategies.py # MACD variants
│ │ └── ... (20+ category modules)
│ ├── indicators/ # Technical indicator library
│ └── mt5/ # MetaTrader 5 integration
│
├── frontend/ # Web UI (HTML/CSS/JS)
│ ├── dashboard.html # Main analysis interface
│ ├── alerts_page.html # Alert management
│ ├── performance.html # Account performance
│ └── ... (20 pages, full Farsi RTL support)
│
├── bot_server.py # Standalone trading bot (port 8001)
├── configs/ # YAML model configurations
├── notebooks/ # Research & demo notebooks
├── tests/ # Test suite
├── models/ # Trained model artifacts
├── data/sample/ # Sample datasets
├── requirements.txt
├── pyproject.toml
└── LICENSE
| Component | Module | Description |
|---|---|---|
| Feature Engineering | ai_core/utils.py |
50+ features: price structure, rolling stats, momentum, volume, time encoding |
| Signal Prediction | ai_core/predictive_model.py |
Ensemble of RF, XGBoost, LightGBM with TimeSeriesSplit CV |
| Regime Detection | ai_core/recommendation_engine.py |
ADX + ATR + slope → trending/ranging/volatile classification |
| Strategy Recommendation | ai_core/recommendation_engine.py |
Performance-weighted scoring with regime affinity |
| Sentiment Analysis | ai_core/nlp_processor.py |
Lexicon-based sentiment + entity extraction |
| Trade Optimization | ai_core/decision_optimizer.py |
Fractional Kelly sizing with volatility adjustment |
| Pipeline Orchestration | ai_core/pipeline.py |
Coordinates all components in a single analyze() call |
| Category | Stack |
|---|---|
| AI / ML |     |
| Backend |     |
| Data |     |
| Trading |   |
| Frontend |    |
Contributions are welcome! Please see CONTRIBUTING.md for guidelines.
This project is licensed under the MIT License — see LICENSE for details.