Trading has always fascinated me — that delicate mix of logic, psychology, and chaos.
But after years of clicking charts manually, I thought: why not let code do the heavy lifting?
So, I rolled up my sleeves, fired up Python, and started building my own algorithmic trading system connected directly to MetaTrader 5 (MT5).
What began as a weekend experiment quickly became an obsession — tweaking indicators, optimizing risk models, and running backtests that ate through my GPU.
How It Started
It began simple:
Connect to MT5 via the Python API.
Stream live data into Pandas & Redis.
Run feature engineering pipelines (VWAP, PPO, SuperTrend, ADX, LSTM signals).
Train and test models with Stable-Baselines PPO & TD3 agents.
Deploy live on my local trading terminal, complete with logs, alerts, and automatic journaling.
It wasn’t pretty at first. I broke things… a lot.
But each failed experiment taught me something — and soon, my bots started making consistent, data-driven trades.
The Results So Far
Here’s a snapshot of the latest live model performances across several pairs:
| Symbol | Status | Win Rate | Total Return | Profit Factor | Max DD | Model Version | Last Updated |
|---|---|---|---|---|---|---|---|
| XAUUSD | LIVE | 71.87% | +500.40% | 4.26 | 1.18% | 4.1_emergency_optimized | 2025-10-01 |
| XAUUSD |  LIVE TRADING | 72.1% | +48.87% | 4.65 | 0.24% | realistic_multi_timeframe | 2025-10-07 |
| US30 | LIVE | 78.6% | +107.8% | 1.34 | 1.97% | 7.0_production_error_fixed | 2025-10-07 |
| EURUSD | LIVE | 81.0% | +17.23% | 3.21 | 0.44% | 3.0_advanced_ppo_22_features | 2025-09-27 |
| GBPJPY | LIVE | 47.1% | +36.48% | 10.0 | 0.39% | 3.0_optimized_balanced_ppo | 2025-09-29 |
| USTEC | LIVE | 50.5% | +57.86% | 1.85 | 16.31% | 3.0_optimized_ppo_30_features | 2025-09-28 |
| EURJPY | READY | 97.8% | +97.8% | TBD | TBD | 267_feature_ensemble | 2025-10-07 |
Each model version represents a new learning cycle — tweaking feature sets, adjusting reward functions, and improving how my agents interpret volatility.
Lessons Learned
Simplicity wins. The more complex the system, the more fragile it gets.
Data cleaning matters more than model selection. Garbage in = garbage out.
Risk control is the true edge. Drawdowns under 2% aren’t by accident.
Patience pays. Most breakthroughs happen after 100+ failed runs.
What’s Next
I’m working on integrating:
A real-time dashboard (Flask + Redis + Power BI) for live monitoring.
A neural agent evaluator that learns from past model behaviors.
Automated retraining triggered by performance drift detection.
Ultimately, my goal isn’t just profit — it’s creating a self-learning, adaptive trading ecosystem that improves with every tick of data.
Let’s Talk Trading + Tech
If you’re experimenting with algo-trading, Python automation, or just love nerding out about data, drop me a line!
I’d love to exchange ideas, compare strategies, or even collaborate on an open-source trading bot project.
#CodenRob #PythonTrading #MT5Automation #AlgoTraderLife