I am Sanman Kadam, a Data Analyst with an MSc Statistics background specializing in quantitative analysis, experimental design, machine learning, SQL analytics, and business intelligence reporting.
I bridge the gap between statistical theory and business decision-making: from designing rigorous A/B tests and early-stopping ML frameworks to engineering production-grade fraud classifiers, customer RFM segmentation, and interactive Power BI dashboards.
My goal is simple: transform complex quantitative data into clear, credible, and decision-ready insights.
| Area | Skills |
|---|---|
| Programming & Libraries | Python, R, SQL, Pandas, NumPy, Scikit-Learn, TensorFlow, SciPy, Statsmodels |
| Databases & Tools | SQL Server, PostgreSQL, MySQL, Docker, Git, GitHub, Jupyter Notebook |
| Visualization & BI | Power BI, Power Query, DAX, Streamlit, Matplotlib, Seaborn |
| Statistics & ML | A/B Testing, Hypothesis Testing, Bayesian Inference, Regression, Classification, Clustering, Time Series |
| Domain | Details |
|---|---|
| A/B Testing & Causal Inference | Power calculations, sample sizing, z-tests, regression adjustment, Bayesian posteriors, HTE |
| Exploratory Data Analysis | Data cleaning, distribution diagnostics, correlation analysis, outlier detection |
| Machine Learning | Classification, regression, SMOTE class balancing, cost matrix evaluation, SHAP explainability |
| Time-Series Analysis | Lag-based forecasting, walk-forward cross-validation, baseline comparison metrics |
| Natural Language Processing | Text vectorization, Sentence Transformers, NMSLIB ANN search, LSTM deep learning |
| Business Intelligence | Star schema data modeling, T-SQL window functions, CTEs, Power BI DAX expressions |
| Degree | Institute | Details |
|---|---|---|
| MSc Statistics | University of Mumbai | Advanced Statistical Inference & Data Science |
| BSc Statistics | University of Mumbai | Core Probability & Mathematical Statistics |
1. Conversion Optimization A/B Testing
Comprehensive A/B testing decision framework for e-commerce checkout flow optimization.
| Attribute | Details |
|---|---|
| Tools | Python, SciPy, Statsmodels, Matplotlib, Makefile |
| Methods | Hypothesis testing, regression adjustment, Bayesian decision framework, HTE |
| Key Outcome | Quantified launch recommendation accounting for controlled user-mix imbalance |
| Focus | Distinguishing true treatment effect from demographic user-mix bias |
| Repository | View Repository |
Structured as a real-world product launch decision, this project evaluates raw conversion rates, adjusts for user characteristics via logistic regression, and quantifies financial impact.
2. Credit Card Fraud Detection
Production-ready machine learning pipeline for detecting highly imbalanced credit card fraud transactions.
| Attribute | Details |
|---|---|
| Tools | Python, XGBoost, SMOTE, Scikit-learn |
| Methods | SMOTE oversampling, cost-sensitive evaluation matrix, modular pipeline |
| Key Outcome | Minimizes business cost (Cost = FP*$10 + FN*$100) rather than naive accuracy |
| Focus | Real-world financial risk modeling with severe class imbalance (0.1% positive rate) |
| Repository | View Repository |
Builds a modular training and inference pipeline using SMOTE and cost-sensitive learning to protect revenue against fraudulent transactions.
3. A/B Test Outcome Predictor
Machine learning framework predicting final A/B experiment outcomes from early-stage partial data.
| Attribute | Details |
|---|---|
| Tools | Python, XGBoost, Random Forest, SHAP |
| Methods | Bayesian features, 1,200 simulated experiments, feature engineering |
| Key Outcome | 90.7% accuracy at Day 5, reducing decision time by 64% (saving 9 experiment-days per test) |
| Focus | Accelerating experimentation velocity without sacrificing decision quality |
| Repository | View Repository |
This project converts experiment monitoring into a supervised classification task, enabling product teams to call experiment winners 9 days earlier than traditional hypothesis testing.
4. Catastrophe Risk A/B Testing
Monte Carlo simulation engine for insurance portfolio risk evaluation and catastrophe modeling.
| Attribute | Details |
|---|---|
| Tools | Python, Streamlit, SciPy, NumPy |
| Methods | 10,000 Monte Carlo iterations, reinsurance layer modeling, KS tests |
| Key Outcome | Tail risk evaluation comparing non-normal profit distributions under catastrophe events |
| Focus | Actuarial pricing, reinsurance excess-of-loss (XOL) impact, and solvency analysis |
| Repository | View Repository |
Simulates catastrophe frequency and severity distributions across 10,000 scenarios, evaluating how reinsurance layers absorb tail risk for insurance portfolios.
5. Retail Sales Analytics Dashboard
End-to-end Business Intelligence project analyzing 18,000+ retail transactions using SQL Server, Python, and Power BI.
| Attribute | Details |
|---|---|
| Tools | SQL Server (T-SQL), Power BI Desktop, Python, Pandas |
| Methods | Star schema modeling, RFM customer segmentation, window functions, interactive DAX dashboard |
| Key Outcome | Interactive executive overview dashboard identifying top 15% SKUs driving 78% of revenue |
| Focus | Full-stack BI workflow from SQL querying to Power BI executive visual reporting |
| Repository | View Repository |
Integrates data cleaning, SQL analytical queries (running totals, repeat customer analysis), and an executive Power BI dashboard.
6. SMS Spam Classification (Machine Learning & Deep Learning)
SMS spam detection system comparing classical machine learning algorithms with an LSTM neural network, deployed via FastAPI.
| Attribute | Details |
|---|---|
| Tools | Python, Scikit-learn, TensorFlow/Keras (LSTM), FastAPI, Docker |
| Methods | Text preprocessing, TF-IDF vectorization, Bidirectional LSTM, API deployment |
| Key Outcome | 98.5% classification accuracy with <5ms inference latency via REST API endpoint |
| Focus | Production model deployment comparing traditional ML and deep learning text filters |
| Repository | View Repository |
Full NLP classification pipeline featuring classical baselines, an LSTM neural network, and a containerized FastAPI endpoint (/predict).
Turning raw quantitative data into clear statistical evidence and decision-ready insights.