Portfolio Optimization RL Enhancement

I need help enhancing my dynamic multistage portfolio optimization model using reinforcement learning (RL). The goal is to maximize returns while managing risk, measured by Mean Absolute Deviation (MAD).

Key Requirements:
- Model involves 4 decision stages: \(t \in \{0,1,2,3,4\}\)
- Initial allocation at \(t=0\) and terminal stage at \(t=4\)
- Investment policy \(\theta\) to be optimized
- Objective:
\[
\max_{\theta} \quad \mathbb{E}\bigl[R_T(\theta)\bigr] \;-\; \lambda \,\text{MAD}\!\bigl(R_T(\theta)\bigr)
\]

Current Issues:
- Code produced incorrect outputs
- Troubleshooting tried on mathematical formulas, RL algorithms, and data inputs

Ideal Skills and Experience:
- Strong mathematics background
- Expertise in machine learning and reinforcement learning
- Coding proficiency, particularly in Python and related libraries (e.g., TensorFlow, PyTorch)

Looking for someone who can identify and resolve the issues in my existing code, ensuring the model learns correctly.