Data Scientist

• Loan Matching(JPMC) - LOB – Trade Matching(Commercial Business)
  This solution helps the internal users to match the Receivables & Wires. With this solution the effectiveness of the user’s increases by approx. 90% and the matching which used to take months and getting stacked, now gets done in the same day & there are no pending matching tasks. Using Lavenstein Distance the string distance is calculated among relavant fields of Receivables & Wires data. Also dollar difference of inflow & outflow amount is checked. And finally a classification model is built to give the probability scores, which says the probability match between the given wire & receivable. Here two separate models are built for agency & trading accounts.
  • Tools - Python, sklearn, AWS Sagemaker
• Compliance Monitoring (JPMC) LOB – Asset & Wealth Management (AWM)
  • Sampling Service for Performance Outlier review
    This solution addresses which group of underperforming accounts/mandates should be investigated by the compliance monitoring officer.
    For dimensionality reduction, predictors are auto-encoded using fully connected neural networks and the dimensionality is reduced. Optimum number of clusters are determined through Sillhouette and clustering is done using DBScan. The results goes through a percentile ranking where a 50 percentile cut off is chosen for sampling. The percentile is done on Composite, Benchmark Z score & Variance. Accounts sampled out are submitted to the compliance monitoring officer for review.
    Local Explainability of individual clusters is implemented using Single Feature Introduction test, which has been implemented from scratch based on a white paper (https://arxiv.org/pdf/1905.09849.pdf) (https://arxiv.org/pdf/1909.13381.pdf)
  • Supervised model to predict underperforming accounts for Performance Outlier review
    The data generated from the above mentioned Sampling Service is used to build a binary classification model – Random Forest, which can predict whether the mandate/account should be investigated by compliance monitoring officer.
• Prior Authorization( Healthcare )
  This project involves automation of the manual tasks conducted by insurance companies & medical staff. Before conducting any medical procedures / examinations / operation’s prior authorization from provider i.e. insurance company is needed to get the payments.
  • NLP techniques used :- NER, Bert
  • Tools – AWS Textract(OCR), AWS ecosystem (sagemaker, lambda, S3, api gateway etc.)
• BIO Age for Vitality Determining Biological age from the output of Bioelectrical Impedance Analysis (BIA) devices. When a person stands on the BIA device, outputs are fed into our model (GBM) after data manipulation and the user is given a Biological Age based on his body characteristics
  • Tools - Python, sklearn, AWS Lambda, Sagemaker
• Loss Control services using Computer Vision(Deep Learning – CNN) This solution hastens the premium generation process by eliminating Risk agent, with premium determination by the underwriter being faster. Solutions built include counting the people wearing hard hats out of those working in warehouse, determining if forklift is driven distractedly, identifying clutters near exit doors & determining if fire extinguisher is out of service.
• Performed end to end implementation of the models starting from collecting data to creating Endpoints (deployment) in AWS platform.
  • Possess good knowledge of Convolutional Neural Networks. Dealt with image classification using Deep Neural Networks and Rekognition API of AWS. Models were created from scratch as well as Transfer learning was used for classification.
  • Architectures Used:- VGG16, Framework:- Keras Tools:- AWS services(Sagemaker, Lambda, S3, API Gateway)
• Call Health (Health Care Domain)
  Call Health Services Private Limited, takes health services to door steps of customers in India. It provides health care services at home as well as diagnostic centers, while providing nursing facilities at home. Performed secondary research and collecting data of the various states of India to identify Global cohorts and descriptive analysis. Using CLINKS data of CH, identified the CH cohorts and identified the key outputs and related them with the Global Cohorts and recommending what further can be done with Call Health Business to improve presence and launching extra services in places having Business potential. Built a product/service level recommendation engine to recommend customers other products. Found bags of correlated Diseases & Business using Market Basket Analysis.
• Private Edge(AIG) Project for an Insurance client, involving calculation of IRPM factor to be used to calculate insurance premium using customer inputs

Assurant - Policy Admin

• It was a P&C commercial insurance product offering its services to Farmers of USA.
• The project had a unique architecture, where implementation is a combination of Multiple LOB’s.
• The uniqueness was that it had the capability of having Monoline as well as Package policy.
• Worked on the Policy Admin – implementing coverage’s, Premium calculation algorithm
• Technology and tools: Duck Creek 4.2.1,TSV Debugger,Trace Monitor,XML Altova spy,XML,SQL Server2008

• P&C commercial insurance product offering services to Farmers of USA.
• Concept of Master and Child policies where certificates were issued in the form of child policies.
• Worked on Batch requirements of forms, automated processes (Renewal, Non-renewal, Cancellation)
• Worked on Migration of policies from Mainframe system to DCT