TIJO KOTTADYIL THOMAS

Agent Activation Project:
• Engineered an XGBoost model with incremental learning that boosted agent activation rates by 35.15%
• Implemented targeted intervention strategy capturing 77.59% of potential activations by focusing on top 30% agents
• Doubled resource allocation efficiency through data-driven agent prioritization
Risk Analytics & Fraud Detection:
• Led model recalibration initiatives for fraud detection model using LightGBM, resulting in improved decile capture and successfully identifying 68% of fraudulent claims upfront, generating INR 93Cr ROI
Cross-Sell / Up-Sell Model Development:
• Created a cross-selling model to identify complementary product opportunities for existing customers
• Implemented a comprehensive model monitoring system to track model drift to trigger alert through AWS SES when re-calibration needed
Customer Analytics and Retention:
• Developed multi-class classification model predicting premium payment patterns through 25 months, improving persistency ratio to 70%.

• Contributed to Quantiphi's Innovation Hub on project QognitionAI, building MLOps pipelines across on-prem cloud
• AWS, GCP, and Azure, and containerized solutions using FastAPI on GCP
• Performed object detection and sentiment analysis experiments using Hugging Face models, and researched a question answering use case for eCommerce, implementing Haystack's retriever-reader architecture for subjectivity handling

Developed a simple web app to predict the marks of a student based on the no of study hours using simple linear regression, HTMl, CSS and Flask