X-Trader
Where Machine Intelligence Meets the Stock Market
A fully automated, end-to-end algorithmic trading platform combining classical technical analysis with a custom-built neural network ensemble — scanning S&P 500 stocks daily and executing live trades autonomously on Google Cloud infrastructure.
S&P 500
Stocks Analyzed Daily
Neural Ensemble
Custom-Built Neural Networks
Google Cloud
Fully Cloud-Deployed
Live Execution
Fully Autonomous Trading
What Is X-Trader?
X-Trader is a private, fully automated algorithmic trading system for S&P 500 stocks. Every trading day it scans the entire index, scores trade candidates using a combination of technical analysis and a custom neural network ensemble, and places live orders through Interactive Brokers — all without human intervention.
The platform covers the entire trading lifecycle: from ingesting real-time and historical market data through multiple data providers, through continuous model training, to live order execution managed by a fault-tolerant, cloud-distributed process stack running on Google Cloud.
End-to-End Automated Trading Pipeline
Signal Generation Pipeline
Technical Analysis Suite
Computes Bollinger Bands, RSI, ATR, MACD, EMA, Stochastic Oscillator, and NATR across the full S&P 500 universe every trading day.
Pattern Detection
Detects candlestick patterns (Hammer, Engulfing) and algorithmically identifies support/resistance levels using line-detection algorithms.
Multi-Factor Scoring
Ranks trade candidates using a weighted priority scoring system, generating long and short signals with computed target and stop-loss levels.
Live Trading Engine
IBKR Integration
Direct connection to Interactive Brokers Gateway API for real-time order placement across live and paper accounts simultaneously.
Pre-Trade Preprocessing
ATR calculation, first-5-minute volume analysis, and gap analysis vs. yesterday's close before any order is placed.
Safety Range Check
Neural network predictions tested across a range of opening price scenarios — fragile entries near the threshold are automatically rejected.
Risk Controls
Per-day and per-position limits, symbol blocklist enforcement, and duplicate instance prevention keep exposure in check.
Google Cloud Infrastructure
Serverless Feature Computation
Google Cloud Functions compute full feature sets on demand — pulling data from multiple market data providers and returning results via HTTP.
Multi-Machine Orchestration
Multiple machines work in concert — each handling a dedicated role in the pipeline. Redis coordinates inter-process communication across the cluster, keeping the system synchronized in real time.
Automated Data Archival
All trading data — predictions, snapshots, socket logs — is synced to Google Cloud Storage automatically after each trading day.
WatchDog Process Orchestration
Fault-Tolerant Monitoring
WatchDog monitors all critical sub-processes, auto-restarts on failure, and enforces strict daily run windows.
Ordered Daily Sequence
Enforces the correct execution order: data collection → preprocessing → live trading — ensuring each stage completes successfully before the next begins.
Desktop Simulator
A full-featured C# WPF application for offline strategy simulation, portfolio P&L backtesting, and broker connection management.
Custom Neural Network Ensemble
At the core of X-Trader is a proprietary ensemble of neural networks, each trained on a different view of the market. Their combined output produces a final trade ranking signal that is more robust than any single model.
Multiple Specialized Networks
Several neural networks are trained independently, each specializing in a different feature space — technical indicators, price patterns, and market context.
Ensemble Aggregation
Predictions from all networks are aggregated into a single confidence score per stock, reducing the impact of any individual model's weaknesses.
Continuous Retraining
Models are retrained regularly on fresh market data, adapting to evolving market conditions and preventing performance decay over time.
Ranking-Based Objective
Networks are trained to rank profitable stocks above unprofitable ones — a custom loss function directly aligned with the trading objective, not a generic classification proxy.
Multi-Source Training Data
Real-time and historical OHLCV data is sourced from multiple market data providers, enriched with computed technical features before feeding into training pipelines.
Built on TensorFlow & Keras
All networks are built using TensorFlow and Keras, with custom architectures developed in-house — including custom loss functions, layer types, and training routines tuned for financial data.
Technology Stack
Languages
Deep Learning
Cloud
Infrastructure