Technology & Architecture
End-to-end system powering real-time AI trading signals with transparent performance tracking and enterprise-grade reliability.
System Architecture
End-to-end pipeline from data ingestion to transparent performance tracking
Data Sources
Multi-Exchange Aggregation
Real-time market data from 15+ exchanges including Binance, Coinbase, Kraken. Order book depth, trade history, and on-chain metrics aggregated via WebSocket connections.
AI Engine
Neural Pattern Recognition
Ensemble of LSTM, Transformer, and CNN models trained on 5+ years of historical data. Generates signals with confidence scores, entry/exit prices, and risk parameters.
Trade Execution
Lightning-Fast Delivery
Signals delivered via Discord webhooks and WebSocket API in <500ms. Redis-backed queue ensures zero message loss with automatic retry logic and delivery confirmation.
PnL Tracking
Transparent Performance
Real-time equity curve calculation with daily PnL snapshots stored in Redis Cloud. Historical performance data accessible via REST API for complete transparency.
Data Sources
Multi-Exchange Aggregation
Real-time market data from 15+ exchanges including Binance, Coinbase, Kraken. Order book depth, trade history, and on-chain metrics aggregated via WebSocket connections.
AI Engine
Neural Pattern Recognition
Ensemble of LSTM, Transformer, and CNN models trained on 5+ years of historical data. Generates signals with confidence scores, entry/exit prices, and risk parameters.
Trade Execution
Lightning-Fast Delivery
Signals delivered via Discord webhooks and WebSocket API in <500ms. Redis-backed queue ensures zero message loss with automatic retry logic and delivery confirmation.
PnL Tracking
Transparent Performance
Real-time equity curve calculation with daily PnL snapshots stored in Redis Cloud. Historical performance data accessible via REST API for complete transparency.
Live data flowing through the system in real-time
Technical Specifications
How Signals Flow Through The System
1Crypto AI Bot (Signal Generation)
The crypto-ai-bot consumes real-time market data from 15+ exchanges via WebSocket connections. Machine learning models analyze price action, order book depth, and on-chain metrics to generate trading signals with entry, stop-loss, and take-profit levels. Signals are published to Redis Cloud channels with <50ms latency.
2Redis Cloud (Message Broker)
Redis Cloud acts as the central message broker with TLS encryption. Connection string: redis://default:***@redis-19818.c9.us-east-1-4.ec2.redns.redis-cloud.com:19818
Signals are stored in sorted sets with TTL for historical queries. Pub/Sub channels enable real-time broadcasting to multiple subscribers. Redis handles millions of operations per second with built-in replication and failover.
3signals-api (FastAPI Backend)
The FastAPI backend subscribes to Redis channels and exposes REST and SSE endpoints:
GET /v1/pnl?n=500- Historical equity curveGET /v1/signals?mode=paper&limit=200- Recent signalsGET /v1/signals/stream- Server-Sent Events for live updates
Signals are validated with Pydantic models and delivered to this frontend and Discord webhooks within 500ms of generation.
4signals-site (Next.js Frontend)
This Next.js 14 app consumes the signals-api and renders live signals with sub-1s latency. SSE connections provide real-time updates to the signals table. P&L charts are rendered with Recharts and updated dynamically. Error boundaries ensure graceful degradation if the API is unreachable. Deployed on Vercel Edge Network for global CDN distribution.
Safety Controls & Risk Management
Position Sizing Limits
Maximum 2% risk per trade with dynamic position sizing based on volatility and account equity.
Stop-Loss Enforcement
Every signal includes mandatory stop-loss levels. Maximum drawdown threshold triggers automatic pause.
API Rate Limiting
Redis-backed rate limiting prevents abuse and ensures fair access to signals across all subscribers.
Performance Monitoring
Real-time tracking of win rate, Sharpe ratio, and maximum drawdown. Auto-disable on threshold breach.
Methodology & Disclaimers
AI Model Training
Our ensemble of machine learning models is trained on 5+ years of historical cryptocurrency market data including price action, volume, order book snapshots, and on-chain metrics. Models are re-trained monthly and validated against out-of-sample data to prevent overfitting. Confidence scores reflect the model's certainty based on pattern similarity and historical accuracy.
Performance Tracking
All P&L data displayed on this site reflects paper trading results unless explicitly marked as "live". Paper trading simulates trades with realistic slippage (0.1-0.3%) and exchange fees (0.1%). Historical equity curves are calculated from actual signal execution timestamps and cannot be cherry-picked.
Risk Disclaimer
Cryptocurrency trading carries substantial risk of loss. Past performance does not guarantee future results. Signals provided by this platform are for informational purposes only and do not constitute financial advice. You are solely responsible for your trading decisions and should never trade with money you cannot afford to lose. Always conduct your own research and consult with a licensed financial advisor before making investment decisions.
Data Accuracy
We strive for maximum transparency and accuracy in all displayed metrics. However, signal delivery latency may vary based on network conditions. Historical data is stored in Redis Cloud with redundancy, but we cannot guarantee 100% data availability in the event of catastrophic infrastructure failure. In rare cases, displayed metrics may be delayed by up to 5 minutes during API maintenance windows.
Frequently Asked Questions
How are signals generated?+
Our AI engine uses an ensemble of LSTM, Transformer, and CNN models trained on 5+ years of historical market data. The models analyze order book depth, trade history, on-chain metrics, and technical indicators to generate signals with confidence scores, entry prices, stop-loss, and take-profit levels.
What is the signals-api architecture?+
The signals-api is a FastAPI backend that consumes live trading signals from Redis Cloud via secure TLS connection. It exposes REST endpoints (/v1/pnl, /v1/signals) and Server-Sent Events (/v1/signals/stream) for real-time signal delivery to this web frontend and Discord webhooks.
How is data streamed in real-time?+
Trading signals are published to Redis Cloud by the crypto-ai-bot. The signals-api subscribes to Redis channels and broadcasts signals via WebSocket and SSE connections. This ensures sub-500ms latency from signal generation to delivery on your screen.
What safety controls are in place?+
Position sizing is capped at 2% risk per trade. Every signal includes stop-loss levels. We monitor performance metrics in real-time and automatically pause signal generation if win rate drops below 55% or max drawdown exceeds 15%. Rate limiting prevents API abuse.
How is P&L calculated and stored?+
Daily P&L snapshots are calculated based on paper trading simulations or live execution results. Equity curve data is stored in Redis Cloud with TLS encryption and exposed via the /v1/pnl REST endpoint. Historical performance is fully transparent and auditable.
What happens if the API goes down?+
The frontend implements graceful degradation. If the signals-api is unreachable, the last known P&L snapshot is displayed with a notification banner. SSE connections use exponential backoff retry logic. All critical data is persisted in Redis Cloud for recovery.
Is the infrastructure scalable?+
Yes. The signals-api is stateless and can be horizontally scaled behind a load balancer. Redis Cloud handles millions of operations per second. The Next.js frontend is deployed on Vercel Edge Network for global low-latency access.
What are the SLA/SLO targets?+
We target 99.8% API uptime, less than 500ms signal delivery latency, and less than 2s page load time (LCP). Redis Cloud provides built-in replication and failover. Monitoring alerts trigger if uptime drops below 99.5% or latency exceeds 1s.
Ready to Experience AI-Powered Trading?
View live signals and performance metrics in real-time.