Building a Multi-Horizon Crypto Prediction System with Transformers
Explore the engineering story of crypto prediction with Transformers. Lessons on transparency and reliability with real metrics.
The project began in January 2026 by questioning the ability to predict Bitcoin's direction. It processed over 87,000 training generations and overcame three major architectural bugs. The hybrid Transformer and BiLSTM architecture provided an effective structure for analyzing financial time series.
The system's accuracy was supported by a verification process based on transparency and independent decisions. However, certain bugs raised doubts about the reliability of the model's predictions. These issues ultimately led to more realistic and reliable results.