Backpack Scalper
SHADOWWebSocket-based scalping bot for Backpack Exchange. Real-time klines + orderbook signal engine, spike detection for ML training data collection.
Key Numbers
At a Glance
Real-Time OB
Data Source
Spike Signals
Detection
Trading + ML Data
Purpose
Backpack
Exchange
Overview
About This Project
A WebSocket-based scalping system built for Backpack Exchange, a newer crypto venue where structural inefficiencies create short-lived trading opportunities. The bot connects to real-time kline and orderbook streams, computing a suite of signals designed to detect momentum spikes and microstructure anomalies.
Beyond active trading, the system doubles as a sophisticated data collection platform. A spike detection module identifies and logs unusual market events -- sudden volume surges, orderbook imbalances, and price dislocations -- generating labeled training data for downstream ML models.
The modular architecture separates feed ingestion, signal computation, and position management into independent components, making it straightforward to add new signal types or adapt the system to other emerging exchanges with similar WebSocket APIs.
Features
What It Does
WebSocket Feed Engine
Persistent connections to kline and orderbook streams with automatic reconnection, message sequencing, and backpressure handling for reliable real-time data ingestion.
Signal Computation Pipeline
Real-time computation of momentum, volume spike, orderbook imbalance, and microstructure signals from raw feed data with configurable lookback windows.
Spike Detection & ML Logging
Identifies unusual market events -- volume surges, price dislocations, orderbook sweeps -- and logs them as labeled training data for downstream ML model development.
Position Management
Automated position entry, exit, and risk management with configurable stop-loss, take-profit, and maximum exposure parameters.
Architecture
How It Works
Challenges
What Made This Hard
Newer exchanges often have thinner liquidity and less predictable microstructure than established venues. Building signals that distinguish genuine momentum from noise in thin orderbooks required careful feature engineering. The system also needed to handle Backpack's specific API behaviors and rate limits while maintaining data collection reliability for ML training purposes.
Stack