Gold’s price is influenced by dozens of factors — from USD strength and bond yields to breaking news. Our goal was to design a model that could predict the price with high directional accuracy and low error, while running in production with live data feeds.
Achievements
• Achieved 96.1% accuracy over rolling 90-day live testing.
• Reduced RMSE to 0.039 via Transformer residuals architecture.
• Implemented 99.9% uptime pipeline with auto-recovery and monitoring.
• Live WebSocket feed delivers predictions to dashboard and API subscribers within milliseconds.
Rolling Accuracy Over Time
What’s Next?
This is just the start. We’re already experimenting with adding live commodity order book data, satellite imagery analysis of mining sites, and real-time central bank announcement parsing to further refine accuracy. The pipeline is modular, so new features can be added without downtime.
If you’d like to see the live dashboard in action, including upcoming macro events influencing our predictions, reach out for access.