At Bharat Electronics Limited, I built an NLP pipeline for information extraction and named-entity recognition from unstructured web text. I designed its collection, deduplication, and text-classification stages to turn raw text into structured, searchable records.
In my fraud detection project, I evaluated a stacked LightGBM model on more than 2 million transactions, achieving a ROC-AUC of 0.96. I also added concept-drift detection with ADWIN and PSI, connecting it to a risk-tiered alert engine and live Streamlit dashboard.
For my Mental Manipulation Detector, I fine-tuned RoBERTa-base for nine-class tactic detection and added SHAP phrase-level explanations. The project includes real-time predictions, a confusion matrix, and 22 PyTest tests in a Streamlit dashboard.
I also analysed business datasets in Python and Pandas during the Tata (Forage) virtual internship, then built Tableau KPI dashboards and performance reports. I'm studying Data Science at Manipal University Jaipur and have also developed cricket analytics projects using ball-by-ball features and IPL auction data.
