Vishwas Kaushik

Actively seeking Data Scientist / Data Analyst roles

Projects

Here are some of my featured projects.

Riskos – Stock Risk Analysis Tool

Problem: Retail investors often lack tools to quantify portfolio risk and forecast market volatility across multiple stocks.

Approach: Developed a full-stack Python + React app with financial APIs to compute VaR, CVaR, and Sharpe Ratio. Integrated ARIMA and GARCH for volatility forecasting across Nifty50 stocks. Containerized with Docker and automated CI/CD via Jenkins.

Impact: Delivered an interactive dashboard where users can instantly analyze portfolio risk and market uncertainty, helping them make more data-informed investment decisions.

Predictive Modeling Case Study Portfolio

Problem: Organizations across domains need reliable ML models for forecasting, classification, and decision-making.

Approach: Completed 11 end-to-end ML case studies in Google Colab, applying regression, classification, clustering, time series forecasting, PCA, CNNs, and cross-validation. Covered diverse datasets in finance, astronomy, e-commerce, and computer vision.

Impact: Built a diverse modeling portfolio showcasing ability to adapt ML methods to different domains, translating raw data into actionable insights.

HR Analytics Dashboard (Employee Attrition)

Problem: High employee attrition increases costs and disrupts workforce stability. HR teams need visibility into turnover drivers.

Approach: Designed an interactive Power BI dashboard (IBM HR dataset, ~1.4K records) with DAX measures to track attrition by department, role, and demographics. Identified low salary, overtime, and dissatisfaction as key drivers; Sales Reps showed the highest attrition (40%).

Impact: Enabled HR teams to design targeted retention strategies by pinpointing high-risk roles and key attrition drivers.

Empowering Rural Healthcare through AI Solutions

Problem: Rural healthcare systems face limited resources for early cancer detection, making accurate triage and diagnosis difficult.

Approach: Built a multimodal pipeline combining medical image models (ResNet18, DenseNet121) with an NLP component analyzing clinical reports. Late-fusion ensembles improved validation accuracy from ~53% to ~84%.

Impact: Showcased potential of AI-assisted diagnosis in low-resource settings, improving prediction robustness and enabling early intervention strategies.

Contact Me