Payal Bhatoa

• Designed a customer–product recommendation system to identify high-propensity LIC products within 120 days post-claim, framing the problem as a scalable customer–product binary classification task.
• Built and deployed multi-model XGBoost architecture (20+ product-specific models) to capture heterogeneous purchase behavior and improve recommendation precision.
• Applied probability calibration techniques to normalize model outputs across products, enabling reliable cross-product ranking and stable customer-level comparisons.
• Developed a ranking and decision framework to generate Top-1 and Top-2 recommendations per customer, integrating business eligibility rules (age, income, policy constraints) to ensure operational feasibility.

• Designed and implemented end-to-end GenAI pipelines using prompt engineering, embeddings, chunking, and Retrieval-Augmented Generation (RAG) across insurance and marketing use cases.
• Benchmarked, evaluated, and selected AWS-hosted LLMs via AWS Bedrock (Claude Sonnet, Claude v2/v2.1, Jurassic Ultra 2) using ROUGE-based metrics, cost, and latency trade-offs.
• Built schema-aligned prompt frameworks to generate structured, deterministic outputs (Excel / document templates) and reduce hallucinations.
• Integrated external knowledge sources, VectorDBs, and retrieval workflows to improve contextual grounding and output consistency.

• Built anomaly detection framework using time-series analysis and clustering.

• Developed sensor optimization and alerting systems integrated with Systrack.

• Designed Customer Engagement Score improving sales productivity.

• Delivered ML-driven insights to improve customer growth and profitability.

• Identified high-potential customers using Logistic Regression generating $10M+ impact
• Deployed NLP-based feedback classification model with 85% accuracy
• Improved usage growth prediction using clustering and XGBoost