I’m a Business Analytics graduate from The George Washington University with hands-on experience building data-driven systems using Python, SQL, and R. My work focuses on applied machine learning, Responsible AI, and analyst-facing decision-support tools, with an emphasis on transparency, interpretability, and real-world usability.

I design and deploy end-to-end analytics solutions that combine interpretable machine learning, statistical modeling, and modern AI workflows. My projects include explainable ML models (EBM, SHAP), forecasting pipelines, bias and fairness evaluations (AIR), and deep learning applications such as CNN-based medical image classification. I have also built LLM-assisted analytical tools that translate model outputs into clear, decision-ready insights for technical and non-technical stakeholders.

With a multidisciplinary background spanning analytics, marketing, and entrepreneurship, I bridge data science and business strategy by translating complex model behavior into actionable insights. I am particularly interested in responsible AI systems that support high-stakes decision environments such as fraud analysis, healthcare, and risk modeling.

Designing interpretable and bias-aware machine learning systems using SHAP, EBM, and fairness evaluation frameworks such as AIR.

Building decision-support tools that translate complex model outputs into clear, actionable insights for analysts and domain experts.

Developing applied AI solutions across domains including fraud detection, healthcare analytics, and risk modeling.

Building scalable data and forecasting pipelines using Python, SQL, and cloud-based data infrastructure.

Bridging data science and business strategy through experimentation, visualization, and interpretable modeling.

Applying the discipline and resilience developed through years of weightlifting and coaching to complex analytical problem solving.

Programming & Data

Python (pandas, NumPy, scikit-learn, statsmodels, PyTorch) • R • SQL (PostgreSQL, SparkSQL)
Data cleaning • feature engineering • data validation • ETL pipelines • exploratory data analysis
Notebook-based research workflows (Jupyter, VS Code)

Business Intelligence & Visualization

Power BI • Tableau • SAS Visual Analytics • ggplot2 • Matplotlib • Seaborn
KPI design • dashboard development • stakeholder reporting • insight storytelling
Data visualization for model diagnostics and decision support

Machine Learning (Applied)

Supervised learning (classification & regression)
Random Forest • XGBoost • LightGBM • Gradient Boosting
SVM • LASSO • Ridge • ElasticNet
Explainable Boosting Machines (EBM)
Deep learning basics (CNNs for image classification) Model training, cross-validation, and performance evaluation
(AUC • ROC • lift • RMSE • confusion matrix)

Explainable & Responsible ML

Global and local model interpretability (EBM, SHAP)
Feature importance and risk driver analysis
Bias and fairness testing (AIR, subgroup analysis)
Transparent model evaluation for audit, governance, and regulatory use cases

Decision Support & Analytics

Alert prioritization and case-level analysis
Score interpretation and threshold analysis
Analyst workflow optimization
Decision-support system design for review-based environments

Forecasting & Statistical Analysis

Time-series forecasting (ARIMA / ARIMAX)
Trend and seasonality analysis
A/B testing • hypothesis testing • statistical diagnostics
Regression modeling and econometric analysis

Applied AI & Analytics Engineering

Streamlit applications for analyst-facing tools
Retrieval-Augmented Generation (RAG) for contextual explanations
LLM integration (LangChain, OpenAI API)
Prompt engineering for structured analytical outputs
Vector databases (FAISS)

Data & Cloud

AWS • Apache Spark
Dimensional data modeling
Scalable data processing workflows
Version control (Git, GitHub)

• Explainable AI Chatbot for Alert Analysis
  Built an award-winning, Streamlit-based decision-support tool using interpretable machine-learning models (EBM) and Retrieval-Augmented Generation (RAG) to help analysts understand key drivers behind alerts, justify model outputs, and prioritize reviews more effectively—reducing manual review time by ~30%.

• Breast Cancer Image Classification with CNN + LLM Summarization Developed a convolutional neural network (CNN) using the BreastMNIST dataset to classify malignant vs. benign breast tissue images and evaluate model performance using accuracy, ROC-AUC, and confusion matrix analysis. Integrated an LLM-based component to generate clinician-friendly summaries of model results, translating technical outputs into interpretable insights for research and healthcare audiences. The repository includes the full codebase and a PDF notebook with outputs for reproducibility.

• Capital Bikeshare Demand Forecasting
  Developed LASSO and Ridge regression models to forecast daily bike demand, supporting operational planning and improving resource allocation accuracy (~85% R²).

• Risk & Forecasting Analytics Project
  Built a forecasting framework in R using ARIMAX models and macroeconomic indicators (e.g., unemployment, housing prices) to support scenario analysis and stress testing in regulated environments.

• Fair Lending & Bias-Aware Modeling
  Developed interpretable models using EBM and SHAP to evaluate and remediate bias in decision outcomes, balancing model performance with transparency and fairness considerations.

• Iowa Liquor Sales Market Entry Analysis
  Built an end-to-end SQL and Python analytics pipeline on AWS with Power BI dashboards to identify demand patterns, underserved regions, and growth opportunities for retail expansion.

LinkedIn https://www.linkedin.com/in/maryamshahbazali/

GitHub: https://github.com/tsjmaryam

Portfolio: https://tsjmaryam.github.io/