At IIT, Roorkee, I engineered Python and Pandas ETL pipelines to clean and normalize 10K+ unstructured environmental records. I also designed interactive Matplotlib and Seaborn dashboards that improved reporting efficiency by 30%.
I applied statistical analysis and hypothesis testing to validate findings and provide insights that influenced research outcomes.
For my credit risk scorecard project, I built SQL feature-extraction pipelines for a 2.6M-loan dataset and benchmarked Logistic Regression, Random Forest, and XGBoost models. The resulting system achieved an AUC of 0.71 and KS of 0.30, with a FastAPI inference endpoint and Streamlit dashboard.
I also developed an NLP-based resume analyzer and fine-tuned a Kumaoni chatbot. The resume analyzer reached 80% semantic parsing accuracy, while the chatbot's LoRA approach reduced trainable parameters by 90%.
