Jun 2024 – Apr 2025 · 10 mos · Bengaluru, Karnataka, India · Hybrid

• Improving customer support by reducing chat/call transfers, callbacks, and repeat contacts, focusing on analyzing chat and call transcripts to pinpoint core issues behind escalations. Using an unsupervised approach, I first generated embeddings with sentence transformers and then clustered data with HDBSCAN to reveal key topics behind transfer problems. BERTopic modeling extracted topics from each cluster, while large language models (LLMs) provided relevant labels for these topics through instruction finetuning with zero-shot and few-shot learning. To manage vast data volumes, I initially built pipelines on sample data and then scaled up by creating a supervised multi-class classifier.
• This involved fine-tuning a prebuilt BERT model by adapting the final layer architecture to accommodate multiple classes, using customized weights based on the LLM-labeled clustered data. This supervised model was then used to tag the remaining data, automating the categorization process. This comprehensive approach enabled proactive issue resolution, improved first-contact resolution rates, and significantly enhanced the overall customer support experience.

Jul 2023 – Jun 2024 · 11 mos · Kolkata, West Bengal, India

• Delivered business-tech insights in insurance by analyzing diverse RFP responses, crafting insightful dashboards thus improving future RFP responses. Also, deployed an Angular App for streamlined access, enhancing data utilization.
• Processed 50 GB of policyholder data, achieving a 10x size reduction10x size reduction via Parquet transformation.
• Employed pandas, scikit-learn, matplotlib, and seaborn to efficiently analyze, visualize, and extract insights from expansive datasets.
• Contributed & enhanced customer segmentation model for targeted marketing. Utilized advanced data analysis tools including pandas, scikit-learn, and TensorFlow, resulting in a 25% increase in campaign effectiveness through data-driven insights and optimized marketing strategies.
• Optimized operational efficiency by creating an automated claims review workflow that intelligently directed claims to relevant teams based on their severity,led to a notable reduction in processing times by 30%.
• Employed & integrated machine learning algorithms to accurately identify potentially fraudulent claims, resulting in a significant reduction of false positives by 20% and substantial cost savings.
• Contributed by developing a predictive model that evaluated risk factors & forecasted claim probability, boosting underwriting precision by 15% and streamlining risk assessment procedures.
• Working closely with the TCS BaNCS Pre-Sales Team, understanding the requirement of the customers to develop P&C and Health Products for demonstration.
• Bagged 5th position in the GenAI Experience challenge hosted within TCS.

Jul 2022 – Jul 2023 · 1 yr · Kolkata, West Bengal, India

• Working closely with the TCS BaNCS Pre-Sales Team, understanding the requirement of the customers to develop P&C and Health Products for demonstration.
• Segregating the Policies issued from various region as per client given records and performing statistics
• and finding out insights for the same.
• Task in making the P&C and Health Product more enhanced and market ready depending upon the
• needs of various regions, worldwide.

Jul 2021 – Jun 2022 · 11 mos · Kolkata, West Bengal, India

• TCS BaNCS Insurance Digital Platform: Front-End Development, Core TCS BaNCS System
• Client: TCS Internal Product Development
• From the scratch development, testing and bug fixing of TCS Bancs Insurance product which includes
• Dashboard, New Business Issuance and Claim section for end-point customers, a complete end to end
• Insurance requirement any Insurance Company to manage its flow.
• Using the Core BaNCS system to develop a DEMO environment of P&C and Health products of TCS
• BaNCS, to meet the requirements for end-point customers.
• POC: Collecting and Converting Data from Oracle DB to Parquet file format and performing analytics
• Used the cx_Oracle library to extract the data from the Oracle Database and then using the Spark to
• convert the entire DB to Parquet file format.
• Using the Pandas library, the Parquet file was converted into a DataFrame.
• Basic charts and graphs were then derived from Parquet file using the Matplotlib, Seaborn and Pandas
• library in order to get the insights on the behavior of the Policy Holders.

Dec 2022 – Jan 2023 · 1 mo · Bengaluru, Karnataka, India · Remote

• Project Overview:
• The goal of this project was to develop a cloud-based insurance premium predictor using AWS, GitHub Actions for Continuous Integration/Continuous Deployment (CI/CD), and Streamlit for creating a web application. The model would predict the insurance premium for a given customer based on various features . This model would be used by insurance companies to provide instant quotes to their customers and help them make informed decisions.