Debosmita Chatterjee

Policy lapses are a critical challenge in the life insurance industry, leading to revenue loss and disrupted customer relationships. Built ML models to predict policy lapses using customer and policy-related data. Experimented with Random Forest and XGBoost variants, achieving 85%+ accuracy and optimized feature selection using RFE to reduce model complexity by 40%.

Identified potentially fraudulent auto insurance claims using machine learning techniques. Processed 10,000+ insurance records with comprehensive EDA through Dataiku DSS. Achieved 92% precision in fraud detection while maintaining low false positive rates, potentially saving significant claim costs.

Interactive Streamlit app for analyzing, forecasting, and comparing stock market data using ML and time series models. Implements technical indicators, clustering algorithms (KMeans, DBSCAN), and statistical forecasting. Processes real-time data from multiple stock exchanges with automated feature engineering pipeline.

End-to-end Multimodal Generative AI application that converts uploaded images into narrated audio stories. Combines computer vision, large language models, and text-to-speech synthesis. Deployed using Streamlit and Hugging Face Spaces, generating creative narratives in under 5 seconds with seamless audio synthesis.

AI-powered virtual insurance assistant that helps policyholders find answers and navigate policy actions. Built with RAG architecture using FAISS vector search and intent extraction. Handles 50+ policy document types with 95%+ accuracy in intent classification, automatically escalating delays beyond SLA thresholds.

Hybrid NLP-powered Flask API that classifies customer service requests and extracts relevant entities using ML, spaCy NER, and regex rules. Deployed on AWS ECS Fargate with Swagger documentation for seamless integration. Processes 1000+ requests/hour with sub-100ms latency using containerized serverless architecture.