System Architecture

This document provides a comprehensive overview of the system architecture for the Signal trading platform, covering the core AI decision-making pipeline, mathematical analysis engine, service-oriented design patterns, and micro-frontend architecture. For detailed information about individual trading components, see Trading Components. For agent system implementation details, see Agent System Reference.

Overall System Architecture

The Signal platform implements a sophisticated three-layer architecture combining AI-driven decision making, mathematical analysis, and distributed frontend components.

High-Level System Overview

Three-Stage AI Trading Pipeline

The core decision-making system implements a three-stage AI pipeline that mimics professional trader thinking patterns, with each stage serving a specific function in the trading workflow.

AI Decision Pipeline Architecture

TTL System and Dynamic Intervals

Each stage implements adaptive Time-To-Live intervals that adjust based on market conditions and position profitability:

Stage	TTL Type	Interval	Condition
SignalOutline	BULLISH_TTL	2.5 minutes	position="long"
SignalOutline	BEARISH_TTL	1.5 minutes	position="short"
SignalOutline	NEUTRAL_TTL	5 minutes	position="wait"
CloseOutline	PROFIT_RISING_TTL	2.5 minutes	Profit growing, price rising
CloseOutline	PROFIT_FALLING_TTL	1.0 minutes	Profit declining, price falling
CloseOutline	LOSS_RISING_TTL	5 minutes	Loss shrinking, price rising
CloseOutline	LOSS_FALLING_TTL	5 minutes	Loss growing, price falling
RiskOutline	CHECK_TTL	15 minutes	Global condition assessment

Mathematical Analysis Engine

The system employs six specialized mathematical services that provide comprehensive technical analysis across multiple timeframes, each optimized for specific trading strategies.

Mathematical Services Architecture

Service Specializations and Parameters

Each mathematical service implements specific indicator configurations optimized for its timeframe and trading strategy:

Short-term Scalping (15-minute candles):

• Data: 144 candles spanning 36 hours
• Fast indicators: RSI(9), StochRSI(9), MACD(8,21,5)
• Tight Bollinger Bands(10,2.0) for precise signals
• Quick moving averages: EMA(8), EMA(21)

Swing Trading (30-minute candles):

• Data: 96 candles spanning 48 hours
• Balanced indicators: RSI(14), StochRSI(14), MACD(12,26,9)
• Standard Bollinger Bands(20,2)
• Medium-term averages: EMA(13), EMA(34), SMA(20)

Position Trading (1-hour candles):

• Data: 48 candles spanning 48 hours
• Comprehensive suite: RSI(14), MACD(12,26,9), Signal(9)
• Long-term averages: SMA(50), EMA(20), EMA(34), DEMA(21)
• Trend strength: ATR(14/20), Momentum(10), ADX(14), CCI(20)

Service-Oriented Architecture

The system implements a sophisticated dependency injection container with service-oriented design patterns, ensuring loose coupling and high testability.

Dependency Injection Container

Service Registration Pattern

The DI container follows a consistent registration pattern for all services:

// Core service registration example
container.bind<SignalLogicService>(TYPES.signalLogicService).to(SignalLogicService);
container.bind<BinanceService>(TYPES.binanceService).to(BinanceService);
container.bind<SettingsConnectionService>(TYPES.settingsConnectionService).to(SettingsConnectionService);

// Math services registration
container.bind<ShortTermMathService>(TYPES.shortTermMathService).to(ShortTermMathService);
container.bind<SwingTermMathService>(TYPES.swingTermMathService).to(SwingTermMathService);
container.bind<LongTermMathService>(TYPES.longTermMathService).to(LongTermMathService);
Copy

Micro-Frontend Architecture

The platform implements eight independent React applications, each serving specific business functions with autonomous deployment capabilities.

Frontend Application Structure

Application-Specific Features

Each micro-frontend serves distinct functional requirements:

Application	Port	Primary Features	Key Components
about-app	3000	System information, project details	About pages, contact info
candle-app	3000	Candlestick analysis with charts	MainView, CoinView, DataView, lightweight-charts
chat-app	3000	Multi-agent AI consultation	MainView, Agent Swarm integration
kpi-app	3000	Trading performance dashboard	MainView, revenue metrics, cryptocurrency-icons
news-app	3000	Cryptocurrency news aggregation	MainView, CoinView, social signals
setup-app	3000	System configuration interface	MainView, SetupView, StateView, ExportView
signal-app	3000	Trading signals management	MainView, signal monitoring
strategy-app	3000	Strategy analysis and backtesting	MainView, CoinView, performance analysis

Agent Swarm System

The platform implements a multi-agent AI consultation system with specialized cryptocurrency consultants coordinated through a central triage mechanism.

Agent System Architecture

Dynamic AI Model Selection

The system implements sophisticated AI model selection through the InferenceMetaService, supporting multiple providers:

Available AI Providers:

• HuggingFace: openai/gpt-oss-120b, DeepSeek-R1, DeepSeek-V3
• xAI: grok-3, grok-3-mini
• Ollama: gpt-oss:120b, deepseek-r1:671b, deepseek-v3:671b

The getActiveInference() method checks Boolean flags in SettingsConnectionService to determine the active model, with automatic fallback to the default model.

Configuration and Feature Management

The system implements enterprise-grade configuration management with Redis-based storage, feature flags, and sophisticated UI-to-backend mapping.

Feature Flag Architecture

Configuration Categories

The system manages three primary configuration categories:

Signal Strength Control (5 levels):

• UI: Single slider (1-5)
• Backend: 5 Boolean flags (isSignalStrengthMinimal through isSignalStrengthMaximal)
• Integration: SignalCoreService.CHECK_SIGNAL_RANGE_FN validates TP/SL thresholds
• Ranges: TP 0.5%-3.0%, SL 0.25%-1.5%

AI Behavior Feature Control (6 filters):

• Signal Risk Filter: Trend coherence requirement
• Signal Volume Filter: Volume growth requirement
• Signal Overlap Filter: Position conflict protection
• Close Risk Filter: Bull market holding logic
• Close With Profit Filter: Early profitable closure
• Close With Loss Filter: Early loss-cutting closure

LLM Selection Control (8 models):

• UI: Dropdown selection
• Backend: 8 Boolean flags (isInferenceSelection0 through isInferenceSelection7)
• Integration: InferenceMetaService.getActiveInference() provides dynamic model selection

Data Storage and Persistence

The system uses a hybrid storage approach with MongoDB for time-series trading data and Redis for configuration and caching.

Storage Architecture

The architecture demonstrates enterprise-grade design patterns with clear separation of concerns, comprehensive dependency injection, and sophisticated configuration management. The three-stage AI pipeline provides professional-level trading decision making while the micro-frontend architecture ensures maintainable and scalable user interfaces.