This paper explores the application of advanced machine learning techniques, specifically Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models, in forecasting US Treasury bond yields. By integrating key economic indicators and policy changes, the study aims to enhance the precision of yield predictions. The research demonstrates that LSTM models outperform traditional RNNs in capturing temporal dependencies and complexities in financial data. The inclusion of macroeconomic and policy variables significantly improves the models' predictive accuracy. The study also addresses the challenge of handling situations where rate hike expectations have already been priced in by market sentiment, using an LSTM model that incorporates market sentiment effects. The results show that the LSTM-based model (CANOAK) performs better in accounting for complex, non-linear relationships in economic data, particularly in scenarios where short-term fluctuations in indicators do not directly translate into proportional changes in bond yields. The study concludes that the LSTM model can achieve a Mean Average Error (MAE) below 2% for a 250-day learning period, making it a superior tool for predicting US Treasury bond yields.This paper explores the application of advanced machine learning techniques, specifically Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models, in forecasting US Treasury bond yields. By integrating key economic indicators and policy changes, the study aims to enhance the precision of yield predictions. The research demonstrates that LSTM models outperform traditional RNNs in capturing temporal dependencies and complexities in financial data. The inclusion of macroeconomic and policy variables significantly improves the models' predictive accuracy. The study also addresses the challenge of handling situations where rate hike expectations have already been priced in by market sentiment, using an LSTM model that incorporates market sentiment effects. The results show that the LSTM-based model (CANOAK) performs better in accounting for complex, non-linear relationships in economic data, particularly in scenarios where short-term fluctuations in indicators do not directly translate into proportional changes in bond yields. The study concludes that the LSTM model can achieve a Mean Average Error (MAE) below 2% for a 250-day learning period, making it a superior tool for predicting US Treasury bond yields.