Apoorv Agrawal

Project - Dynamic Risk Assessment(DRA):

Dynamic Risk Assessment(DRA) is the analytics component of the Intelligence-Led Financial Crime Risk Management(ILFCRM) program built in collaboration with Google using GCP and TensorFlow Extended Framework.

• Working as a member of Research & Analytics function within Risk & Compliance Department which provides research and data-driven risk management via subject matter expertise, Big Data analytics and Artificial Intelligence solutions across lines of businesses and geographies within the bank.
• Developed supervised machine learning model which incorporates expansive set of characteristics and measurable properties related to financial crime risk, referred to as features, that not only generates a single risk score but presents the underlying financial concerns (e.g. money laundering, terror financing, etc.) expressed as both risk probability and descriptive narration that describes the principal reasoning.
• Performed R&D on available data to come up with new and enhanced analysis of data; leverage advanced statistical/ math modelling where required to provide future state analysis and participate in proactive strategy building
• Provided comprehensive analysis and recommended solutions to address complex business problems and issues using data from internal and external sources and applied advanced analytical methods to assess factors impacting growth and profitability across product and service offerings.
• Developed quarterly roadmaps based on impact, effort and test coordination, working with stakeholders to achieve short-term and long-term goals.
• Scaled analytical capabilities across all business areas, evolving analytics to influence bank's strategic planning and executives' decision-making.
• Tested and validated models for accuracy of predictions in outcomes of interest.
• Applied loss functions and variance explanation techniques to compare performance metrics.
• Developed polished visualizations to share results of data analyses. Tech Stack Used: Python, GCP, Tensorflow, DBT, Hive, GitHub.

Project 1 - Predictive Analytics for Engine Program:

• Managed and coordinated an IOT based predictive analytics project for Line Replaceable Units (LRUs) of Pearl-15 business jet engine in Python using pandas, scikit-learn, numpy and scipy.
• Led the team of 8 members including data scientists and data engineers and help them maintain their quality of service by reviewing their work and providing constructive feedback.
• Deployment of the complete analytics module consisting 90 Diagnostic Networks on a cloud platform for continuous monitoring of LRUs.
• Managed the development and deployment program for the project including event planning, resource assignment, and business user communications.
• Implemented anomaly detection methods and predictive analytics for 13 Line replaceable units (LRU) of most advanced business jet engine using EWMA, PCA, GMM and HMM algorithms in an agile environment.
• Carried out extensive exploratory data analysis, collecting various component failure methods from Unit specialists, Subject Matter Experts and Service Engineers.

Project 2- Computer Vision for Damage Detection in Fan Blades:

• Developed CNN architecture-based Deep Neural Network Computer Vision model (using Keras and Tensorflow) for multi-class image classification to perform labour-intensive task of damage inspection on engine parts.
• Executed multi-class classification with 1000+ accurately labeled images of fan blade using VGG-16, ResNet-50,Inception V3, and Inception-ResNet V2 with pre-trained weights using transfer learning on GPU (Nvidia GeForce RTX 2060) using TensorFlow.
• Extracted features by adding a fully connected layer with 512 neurons before the softmax layers of of the best-performing model based on Inception-ResNet V2, which achieved a state-of-the-art classification accuracy of 86.24%.
• Bench marked the model on a public test case (NEU surface defect detection) achieving 96% accuracy.
• Deployed the model successfully on turbine blades data set achieving 91% accuracy.

• Built order allocation engine using multi class classification algorithms such as Random Forest and XGBoost which inflated average order fulfillment rate by 23% and reduced delivery time by 15%.
• Designed and developed demand forecasting model to support inventory replenishment system which reduced excess orders by 33% .
• Planned, engineered, configured and deployed an image annotation tool for product cataloging using deep learning and computer vision algorithms with 95% accuracy on test data set.
• Employed Python for exploratory data analysis,debugging and generating scripts for automating data extraction and transformation tasks.

• Integrated various data sources (Flat Files, CSV files and databases like Oracle and MS SQL Server) with Tableau to deliver 12 dashboards with around 76 reports. Spearheaded the creation of data models and performance optimized SQL queries for visualizing data collected from Oracle PeopleSoft Campus Solution (ERP System).
• Designed, developed and tested 76 ETL Mappings, Mapplets, Workflows, Worklets using Informatica Power Center 9.1.6.
• Responsible for data engineering functions including, but not limited to: data extract, transformation, loading, integration in support of enterprise data infrastructures – data warehouse, operational data stores and master data management.
• Gained extensive knowledge of Hadoop ecosystem including MapReduce Algorithms, Apache Spark,Spark Dataframe/SQL, Pig, Hive, Scoop, Flume and HBASE.