🚀 Engineering & Data Science Portfolio

A comprehensive collection of technical projects demonstrating end-to-end expertise in Machine Learning, Quantitative Finance, and Full-Stack Data Engineering.

📂 Repository Overview

This repository serves as a centralized portfolio containing six production-grade projects. Each directory represents a standalone application or research pipeline, complete with source code, documentation, and rigorous performance analysis.

🧠 Deep Learning & Time Series

Project	Domain	Tech Stack	Key Impact
Solar Energy Forecasting	Smart Grid / Energy	XGBoost LSTM Optuna	Reduced MAPE error significantly vs. ARIMA baselines; engineered hybrid forecasting models for prosumer consumption/production.
CNN Image Classification	Computer Vision	PyTorch torchvision	Achieved 84.6% Accuracy on CIFAR-10 using a custom 3-layer CNN with adaptive pooling and augmentation pipelines.

🤖 Classical Machine Learning

Project	Domain	Tech Stack	Key Impact
Loan Approval Prediction	FinTech / Risk	Scikit-Learn SHAP Random Forest	Built an automated underwriting system with 99.25% Precision and 100% Recall; integrated SHAP for regulatory explainability.
TV Show Analytics	Data Mining	SciPy BeautifulSoup Requests	End-to-end scraper for 200+ shows; applied Kruskal-Wallis & Robust Regression to debunk "Golden Age" TV myths.

📉 Quantitative Finance & Systems

Project	Domain	Tech Stack	Key Impact
Portfolio Risk Modeling	Quant Finance	R GARCH Quadprog	Implemented Mean-Variance optimization (Markowitz) and Dynamic Volatility forecasting using GARCH(1,1).
UniBooks System	DBMS	MS Access VBA SQL	Designed a normalized relational database with RBAC security and automated inventory tracking triggers.

---

🛠 Technical Deep Dives

1. ☀️ Solar Prosumer Energy Forecasting

• Challenge: Mitigate energy imbalance costs in smart grids by predicting erratic prosumer behavior.
• Solution: Developed a comparative pipeline using Gradient Boosting (XGBoost) and Recurrent Neural Networks (LSTM).
• Highlights:
  • Automated hyperparameter tuning via Optuna (Bayesian Optimization).
  • Implemented 5-fold TimeSeriesSplit cross-validation to prevent look-ahead bias.
  • Artifacts: Full technical report (.pdf) and production-ready Python scripts.
• 👉 View Project

2. 🏦 Loan Approval AI & Fairness

• Challenge: Automate loan eligibility while minimizing financial risk and maintaining interpretability.
• Solution: A Random Forest classifier tuned for high precision in the "Safe-to-Approve" band.
• Highlights:
  • Feature Engineering: Created high-impact ratios (e.g., Debt-to-Income, Asset Liquidity).
  • Governance: Utilized SHAP (SHapley Additive exPlanations) to audit model decisions for bias.
  • Performance: Achieved ROC-AUC of 0.999 on the test set.
• 👉 View Project

3. 📊 Quantitative Risk Engine (R)

• Challenge: Model portfolio risk beyond simple standard deviation in volatile markets.
• Solution: A statistical framework combining Modern Portfolio Theory (MPT) with time-series econometrics.
• Highlights:
  • Convex Optimization: Calculated Global Minimum Variance (GMV) and Tangency portfolios using quadratic programming.
  • Volatility Modeling: Integrated GARCH(1,1) to capture volatility clustering and "fat tails" in asset returns.
  • Backtesting: Rolling-window analysis to validate Value-at-Risk (VaR) estimations.
• 👉 View Project

4. 🖼️ CNN Image Classification (Computer Vision)

• Challenge: Implement a robust vision pipeline from scratch without relying on pre-trained models.
• Solution: Designed a custom 3-layer Convolutional Neural Network (CNN) for the CIFAR-10 dataset.
• Highlights:
  • Architecture: Utilized Conv2d blocks with Batch Normalization and Max Pooling; integrated Dropout to prevent overfitting.
  • Augmentation: Applied random rotations and horizontal flips to improve generalization.
  • Result: Achieved 84.6% Accuracy, with strong performance on mechanical classes (Cars/Trucks).
• 👉 View Project

5. 🕸️ Web Scraping & Statistical Analysis

• Challenge: Validate cultural theories ("Golden Age of TV") using real-world unstructured data.
• Solution: A dual-phase pipeline: Automated Scraper (Python/Requests) + Statistical Inference (SciPy).
• Highlights:
  • Data Engineering: Built a resilient scraper to harvest metadata for 200+ shows, handling retries and rate limiting.
  • Inference: Applied non-parametric tests (Mann-Whitney U, Kruskal-Wallis) to handle non-normal rating distributions.
  • Insight: Disproved "Longer is Better" myths using robust regression analysis.
• 👉 View Project

6. 📚 UniBooks Management System (DBMS)

• Challenge: Replace manual bookstore tracking with a scalable, atomic transaction system.
• Solution: A relational database system built with MS Access and VBA automation.
• Highlights:
  • Schema Design: 3NF Normalized database ensuring data integrity across Inventory, Sales, and Procurement.
  • Automation: VBA triggers for real-time stock level checks (Inventory < Order_Qty logic).
  • Analytics: SQL-driven dashboards for "Best Sellers" and monthly revenue tracking.
• 👉 View Project

---

⚡ Getting Started

Each project is self-contained. To run a specific project:

1. Navigate to the project folder.
2. Read the local README.md for specific dependency installation (e.g., pip install -r requirements.txt or R library installation).
3. Launch the corresponding Jupyter Notebook (.ipynb) or R Script (.R).

# Example: Cloning the repo
git clone [https://github.com/andyhu11/Project-Documentation.git](https://github.com/andyhu11/Project-Documentation.git)
cd Project-Documentation

---

📄 License

This repository is licensed under the MIT License. See individual project folders for specific third-party attributions.

---

📫 Connect with Me

• LinkedIn: Andy
• Portfolio: github.com/andyhu11/Project-Documentation
• Email: jiahuiapply26@163.com