Subhasmita Maharana

About

Data/ML practitioner with hands-on experience building NLP systems from prototype to production: delivered a Twitter sentiment classifier with robust preprocessing, SVM modeling, and Power BI reporting, and built entity-resolution pipelines for messy multi-source customer data (reporting ~95% improvement in unique entity identification). Also implemented semantic linking/search using SBERT embeddings with FAISS vector retrieval and domain fine-tuning (reported ~15% precision lift), and applies production workflow best practices (Airflow/Prefect, Docker, Azure ML/Databricks, Great Expectations).

Experience

Education

Key Strengths

• End-to-end NLP sentiment classification pipeline (preprocessing, TF-IDF, model selection, tuning, evaluation)
• Strong model selection and validation practices (GridSearchCV, k-fold CV, precision/recall tradeoffs)
• Entity resolution for messy multi-source customer data using blocking + fuzzy/phonetic matching
• Quantified improvement: ~95% accuracy improvement in unique entity identification
• Semantic search/data linking using SBERT embeddings + FAISS vector similarity search
• Fine-tuned SBERT with quantified gain (~15% linkage precision) and relevance evaluation (MRR, top-k accuracy)
• Production-grade data workflow engineering (Airflow/Prefect orchestration, testing, logging, Docker, cloud tooling)
• Data quality enforcement with Great Expectations and schema/range checks
• Communicates results via BI and visualization (Power BI dashboards, t-SNE clustering visualization)