Vikram Kalister

Built a causal inference pipeline using synthetic control methods to estimate the effect of a 1988 California tobacco policy on cigarette sales. Validated results using placebo testing across 38 comparison states and MSPE-based model fit filtering.

Fine-tuned a PyTorch-based YOLOv8 model to 89% mAP@0.5 across four damage classes on a 5,000-image augmented dataset. Containerized an XGBoost cost estimator with SHAP breakdowns in Docker, exposed via a FastAPI REST API with a Streamlit front-end.

Achieved 85%+ recall on a customer churn task by training a SMOTE-balanced, hyperparameter-tuned XGBoost classifier on 7,000+ telecom records. Built a tenant screening and retention prediction platform using Python, Streamlit, and ensemble ML algorithms to automate rental-property risk decisions.

Built an agentic LLM assistant integrating OpenAI API and LangChain to provide natural-language data insights. Engineered a modular preprocessing pipeline with automated outlier detection, reducing data prep time by 70%.

Improved credit report data accuracy by 20% across a 1M+ customer database by writing SQL validation scripts. Built 10+ Power BI dashboards tracking Microsoft Teams call-volume KPIs, replacing weekly manual reports with self-serve analytics.

Pursued a Bachelor of Science in Data Science with coursework in Advanced ML, Data Visualization, and Big Data Processing.

Completed a Bachelor of Science in Business Administration with a thesis on crisis management in the bank and credit industry.

Completed a Bachelor of Science in Business Administration with a thesis on crisis management in the bank and credit industry.