Projects

Document Anonymization :According to the GDPR, personal data must be protected by both private companies and public bodies. In the event of a violation of the GDPR, there may be a risk of high fines in addition to loss of reputation. Client want to anonymized these entities to overcome this issue. Developed the solution using OpenCV, NLP and Yolov5 which can anonymize entities in documents which is in German language. 

Document Data Extraction 

Phase1 - Copies of customer identification and address verification are collected from the customers. These documents are been checked manually and then updated on the system. As this process is very time taking ,we developed the solution which will process the documents such as Passport and Driving License extract the data from these document using OpenCV ,EasyOCR and Tesseract and then can be updated in the system in one go. This will help to reduce both cost and time. 

Phase2 - There are Valuation report associated with each mortgage account. These Valuation report also been scrutinized by bank representative manually. The Bank representative looks for specific fields from the each page of the valuation report. As this process is time taking we have developed AIML based solution which can extract the data from these valuation report and also do field level extraction using Easyocr. 

Predicting Server Load 

The main objective of this project is to predict the response time of the load happening on the server to avoid server breakdown during Performance Testing. We have developed Regression model which will use the predictor variable like CPU utilization, Memory available, transaction count and Maximum users at specific instance and will predict the response time. This will help the Testing team to make the infra ready for the stress testing of the application and avoid application breakdown.

HSBC.ai

 HSBC.ai is a platform built for citizen data scientist who can build and do prediction on their dataset with very minimal Data Science knowledge. We have developed API which will cater all the module of Machine learning model like Data Loading, Data Exploration, Data Cleaning, Text Cleaning, Data Scaling, Model Building, and Model Prediction. We have tried to build a platform similar to H2O flow. 

AutoHealing - MLOps 

AutoHealing Web based AIML solution which will automatically handle the system generated incidents using Incident classification model. Aditya Sharma 

Incident Classification

As we have already built the model using clustering algorithm to cluster the similar incidents. However later when a new data comes in, then running the model with the full data becomes a challenge as the data size increased drastically with time. So we developed the Classification model using Clustering output, so that the model can be run on Delta data rather than on full data to save execution time. 

Incident Analytics using NLP

Building NLP based solution that will cater the client to identify the frequent incident and their resolution which are similar across the data and cluster them together using DBSCAN. This will help the client to identify and eliminate unnecessary incident to occur in future and thereby reducing manual intervention. 

Trade Prediction and Trade Limit Suggestion using Time Series 

The objective of this project is to extract data from the Data base of the client ,create data frame using important features and build a Time series model that can be used to predict the trade pattern for every hour of the client using historical data. Also a feature has been added to suggest the limit to which client can trade on business days. Along with Model we have developed script which will tell each client historical trade behavior by calculating Trade value and Trade pattern of the last day, last week, last month and last quarter. 

Entity Matching using NLP

The objective of this project is to extract data from different database tables and build a solution that will help to identify Entities (Org Name, Region , State,Domain) from different sources which are similar and create a master data which will have all the Entities across all the sources which are similar using Fuzzy Matching. This will help the client for Master data Management. 

Customer Credit Limit Assurance 

There are many customers who wanted to increase the credit limit of their credit card. However bank needs to identify if the customer is worthy of been given high credit limit services by analyzing various features using his past historical data followed by Machine Learning Classification Model. This will help the bank to identify the credit worthiness of the customer and help increasing the credit limit in no time.

Global Risk Solution

The objective of this project is to perform data analysis and create dashboard using Eagle pace tool(Data Analytics Tool of BNY Mellon).This will help to identify how the portfolios are performing in the given period of time(Weekly, Monthly, Daily). Verifying the reports using Sql queries.

SBA One - Defaulter Prediction 

SBA One platform is one of the loan application platforms of the major bank in US where this platform provides loans to small businesses. The objective of this project is to extract all the information of the Customer seeking for Loan from the SBA one platform and build Machine Learning Model to predict which customers can be default to avoid giving loan to such type of customers. 

Credit Card Approval Automation 

The objective of this project is to identify customer worthiness using historical data and based on the credit worthiness classify the customer high, medium and low risk customer. This will help the bank to identify the each segment of customers so that bank can provide credit cards as per the customer history and usage.