• Document Automation: Developed full workflow of information extraction for an enterprise’s system,optimizing document processing with AWS Textract OCR, and implementing a state-of-the-art modelBERT-powered name entity recognition (NER) systems. Reduced manual workload by 60 hours weekly.• Conducted exploratory research with large language models (LLM), including Falcon-7b and Llama-27B-chat, to improve information extraction performance with advanced prompt engineering strategies,leveraging Generative AI techniques. Outperforming current BERT-based NER systems in the first iteration.• Risk Mitigation Analysis: Created a machine learning classification model to identify high-litigation-risk cases, resolving a $17 million issue for the company.• Applied a variety of classification algorithms, including Random Forest, XGBoost, AdaBoost, SVM, andLightGBM to address the complex classification, achieving an accuracy of 86% with the hyperparameter-tuned model• Implemented Explainable AI techniques such as LIME and SHAP to enhance model interpretability andtransparency in showcasing modeling results. Explored model deployment (MLOps) tools for integrating the predictive model into the workflows.POC - Premium Leakage Detection | Python, Neo4j Cypher, Neo4j GDS• Converted a complex relational database into a Knowledge Graph and applied advanced graphML algorithms, harnessing Neo4j and Neo4j GDS library. This led to the identification of $5 million in anomalies within historical data, significantly improving fraud detection.POC - Semantic Similarity | Python, NLP Transformers, BERT, FAISS• Developed AI to estimate reserve values, enhance operational efficiency, and prevent legal charges, reducing search time from 1 hour to 8 seconds.• Utilized Sentence Transformers to convert categorical columns into embeddings and employed FAISS (Facebook AI Semantic Search) for optimized querying, achieving an 84% accuracy rate.