The traditional methods of portfolio optimisation though perform well, yet at times, they fail to understand the changing market situations in the real world and can have issues in handling assets that are correlated. Though the Hierarchical Risk Parity (HRP) represents an advancement in the portfolio optimisation methods, the performance of HRP could be further improved by integration with deep learning techniques. This project investigates if the combination of advanced deep learning architectures with HRP can improve portfolio optimisation while maintaining interpretability and practical implementability. First, an integrated framework is developed for HRP enhancement using two distinct deep learning models, Long Short-Term Memory (LSTM) and Transformer. Three new hybrid models have been developed and evaluated: the transformer-HRP model and two variants of LSTM-HRP, one deep and the other shallow. The dataset chosen has 50 stocks across the different categories for 20 years, from 2004 to 2024. The dataset undergoes preprocessing first, following which all three models are designed. After which their performance is evaluated thoroughly using a framework consisting of various metrics. The main contributions and achievements of this project are:

• Development and implementation of three new integrated models that are able to com bine machine and deep learning techniques with HRP while maintaining interpretability.

• Creation of a preprocessing pipeline for financial data that effectively handles both deep learning and HRP optimisation requirements.

• Selection of appropriate frameworks for built-in evaluation that include the theoretical performance of the developed models and actual implementation issues.

• Recognition of the significant trade-offs that could exist between model sophistication and realism for model applications which show that sometimes more basic models are more realistic as exhibited by a higher real-world performance of the shallow LSTM-HRP model compared to the profound deep LSTM-HRP model.

• It can be inferred from this project that deep learning enhancements can strongly im prove the results of portfolio optimisation but with special attention to implementation constraints.

This work helps to enrich the developing portfolio management field and shows how deep learning techniques can be incorporated into classical portfolio optimisation techniques in a comprehensible and practical way.