Sanket
Kothiya

Predictive analytics platform that evaluates AI hardware performance, correcting manufacturer bias and forecasting GPU efficiency with research-grade accuracy.

• Built Random Forest & XGBoost models reaching 94% accuracy on GPU performance prediction.
• Designed a normalized PostgreSQL schema with 8 interconnected tables.
• Implemented bias-correction algorithms to neutralize manufacturer reporting skew.
• Shipped an end-to-end ETL pipeline feeding a real-time Streamlit dashboard.

Power BI dashboard analyzing attrition across 1,470 employees — surfacing where talent slips through and giving HR a filterable view across department, role, age, and tenure.

• Analyzed attrition with DAX across department, job role, age group, and tenure.
• Engineered KPI measures: attrition rate, active headcount, age-band segmentation.
• Added slicers for filtering by education, gender, and department.
• Layered validation and segmentation logic to keep every view reporting-accurate.

Compared Facebook Ads vs AdWords for a marketing agency — using hypothesis testing and regression to recommend where every ad dollar should go.

• Ran an independent t-test to prove one platform converts significantly better.
• Trained a linear regression model predicting conversions from clicks.
• Added cointegration testing to find long-term spend–conversion relationships.
• Translated stats into a budget recommendation for non-technical stakeholders.

An interactive Streamlit app that benchmarks stocks against the S&P 500 — calculating beta, expected return and normalized price movement on the fly.

• Built reusable functions for normalization, daily returns and beta.
• Pulled live market data via yfinance and merged on trade date.
• Rendered interactive Plotly charts for price comparison.
• Computed CAPM expected return against the S&P 500 benchmark.