Apoorva Joshi Email & Phone Number

Seattle, Wa, Us

Fort Mill, Sc, Us

Armonk, New York, Ny, Us

• Controls Analysis NLP Framework : Led a Data Science team in developing a Natural Language Processing Framework (Topic Modelling, Document Similarity, Document Classification) for analyzing and standardizing regulatory & compliance documents leading to a reduction of 80% in inspection time invested by Auditors. Extracted taxonomies from Control description documents by generating document embeddings using doc2vec, reducing dimensions using UMAP, clustering documents using HDBSCAN, and extracting Topic keywords using class based TF-IDF scoring mechanism. Identified similar and duplicate Controls using Latent Semantic Analysis & Cosine Similarity measure. Classified Issue documents into Principal Risk categories by using TF-IDF, Principal Component Analysis and XGBoost• COVID-19 Chatbot (Web, SMS, Call)Built a chatbot for addressing COVID-19 questions in the Financial services and Insurance domain. Leveraged Watson Assistant, IBM Voice Gateway and Twilio to enable webchat, voice and SMS functionality for the chatbot. Enhanced the quality of the conversation over call by tuning the Speech-to-Text and Text-to-Speech models. • Control Strength and Weakness Assessment Identified similar controls and similar risks using control descriptions and risk descriptions respectively. Performed Latent Semantic Analysis by converting descriptions into TF-IDF vectors and applying Singular Value Decomposition. Calculated control and risk similarity using Cosine Similarity measure. • Operational Risk Category Reallocation and Risk RankingReallocated risks to the appropriate risk categories using risk descriptions. Implemented Natural Language Pre-processing tasks, TF-IDF vectorization and performed supervised classification using XGBoost to classify risk descriptions into their respective risk categories.

Armonk, New York, Ny, Us

• Retail Transformation AnalyticsLed a team in designing, developing and deploying an Analytics based production level application on AWS along with an executive roadmap for a regional credit union institution for optimizing their branch locations and human resource allocation strategies. Solved client’s problem of allocating members to branches by leveraging foot traffic (8.1 TB data processed on PySpark), demographics, client data and by engineering, forecasting (Holt Winter’s and Exponential Smoothing) and optimizing (Linear Optimization) KPIs in R to demonstrate pre and post-transformation states.• IBM Watson Contact Center Recordings Analyzer Led a team in implementing an end-to-end Text-to-Speech Analytics pipeline for a financial institution, that transcribes call center recordings using IBM Streams, processes the transcripts in a Spark environment to provide custom analytics, which are visualized in Kibana using Elasticsearch. The Spark environment utilized NLP Python modules like spacy, textacy and gensim to process raw transcripts and Watson API’s like NLU and Tone Analyzer to generate sentiments and tones to generate insights.• Time Series ForecastingDeveloped production code in R for end-to-end automation of 780 different Time Series models for forecasting expenses of a software company. Utilized imputation techniques like Kalman Filter for filling missing series and R packages like Prophet and Forecast for implementing variations of ARIMA and Exponential Smoothing.

Armonk, New York, Ny, Us

• Worked on a Proof of Concept based on a Third-Party Check Fraud Detection use case that provided a Machine Learning solution to a regional bank for reducing their fraud losses ($1M) incurred due to high false positive rate generated by their rule based approach.• Demonstrated an increase of 89% in precision (without reducing recall) as compared to the approach implemented by the client.• Implemented the solution on a Spark framework, in an agile environment, by testing and analyzing performances of Decision trees, Random Forests and Neural Networks using PySpark’s mllib and Python’s scikit-learn packages.

Houston, Texas, Us

· Provided Data Warehousing and Business Intelligence services to a leading global telecommunications company for running their marketing campaigns and driving their data-driven insights.• Reduced the load times of the Teradata Presentation Layer tables by 5% through query optimization as a part of the System Improvement Plan.• Facilitated the integration, implementation and management of more than 200 ETL processes in the production phase and executed the changes in the business requirements by modifying AbInitio (ETL tool) graphs.• Maintained data quality across all the layers of the enterprise data warehouse tables and the generated reports.• Reduced the number of hours spent by human resources in fixing issues in the production phase by 30% through automation of manual tasks by creating work-around scripts.• Supplied petabytes of good quality data to business critical applications by carrying out ad hoc data loads.

Master Of Information Systems Management, Business Intelligence And Data Analytics

Bachelor Of Engineering, Computer Science