Dhanunjai Ram Guri

Economic Data Visualization Tool Development

• Developed a Python tool capable of generating dynamic dashboards for economic forecasts visualization.
• The tool allows for effortless comparisons for numerous variables across various forecasts cycles and scenarios.
• Enabled updating of charts and tables in an existing PPT template while seamlessly creating new slides with spatial consistency with potential of generating hundreds of slides with charts & tables within minutes.
• Integrated with a user interface(UI), facilitating widespread access for multiple teams including non-technical members within the organization to observe the evolution of economic forecasts with time.

Basel Production

Support annual CCAR submissions and Mid-Year Financial Assessments for the Fee Income and Expense Models for the Personal Loans & Lines Portfolio.

• Data preparation from multiple data sources and relational databases including Ad-Hoc reports, relational databases with transaction history, economic datasets and upstream loss forecasting model.
• Run the forecast engine to produce outputs from 6 statistical models. Analyze key drivers for the models and act as the point of contact to downstream users for all issues ranging from conceptual clarity to output analysis.
• Significantly accelerated the forecast analysis & monitoring PPT processes using Python with 90x efficiency.
• Incorporated PSI4 into the production control process to minimize errors during data preparation.

Customer Attrition Prediction NLP Challenge

A solution to identify the potential customer churn using the customer-bank communication data in any mode.

• Retrieve valuable insights from both customer interaction and bank responses through Word Cloud and Sentiment Analysis, utilizing models such as AllenNLP and RoBERTa.
• Recognized customers with comparable inquiries/ similar characteristics using TF-IDF & cosine similarity.
• Unknown/New topics were identified by utilizing topic modeling (using GENSIM) and the LLM BERT.
• Developed weighted ensemble models with balanced class weights on PCA components of text-transformations.

Research Content Recommender Model

DBS is a leading Singapore Bank. Group Research, Fund Selection Team, CIO teams publish articles periodically on various topics, ranging from global outlook, strategies, and investment instrument recommendations. These articles are available in DBS Public Webpage (PWEB) that can be searched and accessed organically.

Impact: For a given CIO content, find the customers who are more likely to click the content and do relevant transactions after clicking.

Feature Mart (UNO)

UNO feature mart is all about building features to multiple data science projects. Under migration process converted several SAS scripts to Hive SQL.

Giro Anomaly

The Giro Anomaly Model uses both customer specific and biller-specific information and combines all features using a Decision Tree based predictive model to identify anomalous transactions for Intelligent Banking Customers.

L2 - Quick Links Recommendation Model

Quick links are smart shortcuts in the mobile app. Goal of the use-case is to surface them dynamically by customers need on daily basis for both Retail and iwealth Customers. This Recommendation system helps identify the top 10 most likely quick links for each customer.

Everyday Banking - Drive Cashless

Everyday Banking - Drive Cashless model identifies the customers with high ATM withdrawals and encourages them to migrate to online gateway methods using push notifications. Utilized customer and product prioritization scoring techniques to identify customers based on ATM withdrawal and CASA transactions and initiated push notifications.

3D-Point cloud LIDAR

The data consists of the information about the transformers images with consists of assets like GSM and Microwave antenna. model should predict the class with a bonding box around an object.

Corrosion detection

Corrosion is a clear indicator of the state of the hull metallic structures, and, thus, it is of great interest for the surveyor. I have used WCCD algorithm for corrosion detection and VGG16 Architecture for classifying rust vs no rust in images.