> Problem & Impact
Investors navigating volatile markets require more than just basic charts. They need a unified, data-driven platform capable of dissecting gold's complex price dynamics across multiple timeframes and quantifying its interconnectedness with critical macro factors like USD strength (DXY, EURUSD, USDCHF) and broader equity markets (S&P 500). Traditional manual analysis or siloed models often fail to capture crucial regime shifts – whether driven by risk-on/off sentiment or inflation fears. Aurum Vision was engineered to bridge this gap. It provides an end-to-end solution encompassing robust data engineering, advanced feature extraction, state-of-the-art machine learning, and interactive visualization. The platform empowers users to move beyond simple price watching and gain deep, actionable insights into gold's behavior, enhancing strategic decision-making in dynamic financial environments.
> My Role
This project was conceived and executed as a solo full-stack endeavor. I was responsible for the entire pipeline, from ingesting raw financial data and implementing sophisticated data engineering processes to developing advanced ML models, building the backtesting engine, and designing the interactive dashboard interface.
> Architecture & Approach
Aurum Vision's core strength lies in its structured approach to integrating diverse data sources and analytical techniques. The architecture is organized to process multi-timeframe and macro data efficiently:
- Data Ingestion & Loading: A robust Python engine designed to handle historical OHLCV data for gold and key macro factors.
- Feature Engineering: Extracts rich signals from raw data, including technical indicators, multi-timeframe processing, correlation analysis, and fractal analysis.
- Modeling: Implements Hidden Markov Models for market regime identification, neural networks for signal generation, and uncertainty quantification.
- Hyperparameter Optimization: Uses Optuna for advanced tuning, exposed through the Streamlit interface.
- Strategies & Backtesting: Includes rule engines, AI signal models, strategy optimization, and vectorized backtesting.
- Dashboard: Interactive Streamlit interface for data visualization and analysis.
> Key Technologies & Tools
The platform leverages a comprehensive technology stack:
- Languages: Python
- Core Libraries: pandas, NumPy, scikit-learn
- Financial/Quantitative Libraries: pandas-ta, vectorbt, QuantLib
- Machine Learning Frameworks: PyTorch / TensorFlow, XGBoost
- Optimization: Optuna
- Explainability: SHAP
- Interactive Dashboard: Streamlit
- Development Practices: Docker, GitHub Actions
> Prototype Results & Insights
Initial analysis and prototype backtests have yielded valuable insights:
- Cross-Asset Correlations: Quantified gold's historical correlation with key macro factors.
- Directional Accuracy: Transformer prototypes showed potential directional accuracy above random baseline.
- Rule-Based Strategy Performance: Demonstrated promising Sharpe Ratios over historical data.
- Learnings: Developed deep understanding of time series alignment, feature engineering impact, and macro regime integration.
> Future Roadmap
Aurum Vision is designed for continuous evolution:
- Extend data coverage for gold and macro factors through the present.
- Implement rolling window retraining for ML models to adapt to changing market dynamics.
- Integrate real-time WebSocket data feeds for live signal generation.
- Explore order-book and liquidity features for intraday analysis.
- Plan for a more scalable deployment with a React/Tailwind frontend and FastAPI backend.
> Skills Demonstrated
This project showcases a wide range of technical and domain expertise:
- Applied Time Series Analysis & Forecasting
- Advanced Data Engineering & ETL Pipeline Design
- Machine Learning Model Development & Evaluation
- Hyperparameter Optimization & Model Explainability
- Quantitative Finance & Backtesting
- Cross-Asset Analysis & Feature Engineering
- Full-Stack Development (Streamlit MVP, Python Backend)
- Problem Solving in Complex Data Environments
> Project Stats
STATUS
ANALYSIS_ENGINE_ONLINE
CATEGORY
Python, Machine Learning, Data Analysis
COMPLEXITY
> Tech Pulse
> Project Gallery