Soumyajit Sahu

RAG Document Search Application

• Engineered a hybrid RAG pipeline for internal document search combining FAISS semantic vector indexing with BM25 lexical retrieval, selected based on research showing hybrid approaches outperform pure vector search on domain-specific corpora — achieving 75% retrieval relevance across internal PDF documents and serving results via a Streamlit interface powered by Groq LLM.

Retail Customer Analytics and Demand Forecasting

• Built customer segmentation models using RFM-based K-Means clustering on 525K+ retail transactions and applied the Apriori algorithm to identify frequent product combinations, enabling targeted cross-sell strategies.
• Developed SARIMA/Prophet time-series forecasting models for retail demand planning; A/B testing of segment-based campaigns versus random targeting showed 22% higher conversion rate and 18% improvement in cross-sell purchases.

Telecom Churn Case Study

• Conducted EDA on telecom customer data to analyze churn patterns, identifying a 26.5% churn rate and key drivers including contract type, tenure, and monthly charges.
• Built and optimized an AdaBoost churn prediction model using RandomizedSearchCV, achieving 73% accuracy and highest F1-score among evaluated models, enabling proactive retention interventions.