Tapas Bhadra

Data Scientist Intern

• Collaborating with lead data scientists to understand end user sentiment and make recommendations to an e-commerce client using user reviews
• Performing NLP based tokenization, similarity measures like Levenshtein distance on over 2.5 million reviews, implementing pre-processing on text features like N-gram, frequency, POS tags and similarity & assess performance of the model.
• Optimizing client’s search engine results by 29% by creating product embeddings and calculating similarity with keywords

Data Science Intern

• Collaborated with marketing teams to improve business strategy & pricing, analyzed commodity volume and retail price data, using causal inference identified opportunity with projected business revenue of $3.5 million over 1 year.
• Conducted trend analysis on freight cost, shipping time series data to identify customer price sensitivity for 5 products created data visualizations & dashboards.
• Designed database with operations team to automate inventory management procedures across the company.

Teaching Assistant, Data Mining and Predictive Analytics

• Worked with Prof. Gah-Yi Ban and helped with the execution of lesson plans, created and graded assignments.
• Conducted weekly office hours to help students with their doubts in machine learning

Research Assistant

• Evaluated the extent to which websites track users by analyzing third party cookies shared across domains and visualized using Social Network Analysis (SNA) to provide team with valuable insights.
• Assessed similarity between 10 privacy bills using TF-IDF, cosine similarity to understand trends across US states.

Data Scientist

• Led the restructuring of 10+ client-based Salesforce CRM tool with engineering team, conducted requirements specification to incorporate an in-house co-browsing solution to enhance efficiency of Service Reps.
• Worked cross-functionally with customer success team to increase efficiency of 35+ products by analyzing the generated data and customizing the product basis client infrastructure and usage.
• Set and managed timelines, holding internal and external stakeholders accountable while understanding the flexible nature of client deadlines while boosting efficiency in the deployment process by 40%.
• Increased product efficiency by analyzing the product generated data by 70%

Data Science Intern

• Analyzed and extracted features from a database of images containing handwritten texts.
• Used clustering algorithms for classification and delivered a comprehensive comparative study of the accuracy of the algorithms.
• Developed a model using python (NumPy, pandas, TensorFlow), MATLAB to implement K-nearest neighbors, TSNE, and ther clustering algorithms on handwritten digits data and achieved accuracy of 97%

Data Analyst

• Created an information system to include reviews and feedback from various users of the application.
• To improve user experience, developed a machine learning application using Random Forest Classifier and carried product analysis that improved customer retention by 15%

University of Maryland | College Park, MD

GPA: 4.0

Content-based Song Recommender System using Spotify API

Technologies used: Pandas, NLTK, TF-IDF, Cosine Similarity, Spotipy, Python

• Scraped, extracted song data containing 25 features using spotify API, created a dataset containing 2000 tracks from 10 genres
• Created TF-IDF vectors of the song genres and artist information
• Performed feature engineering and normalization
• Calculated Cosine Similarity of the training data and user playlist and recommended 10 songs based on user features

Here is the Github Repo

Kickstarter Data Analysis to predict Successful Projects for Funding

Technologies used: R, Tidyverse, GGplot2, TF-IDF, XGBoost, Rstudio

• Performed exploratory data analysis, normalization, and correlation analysis using R to select relevant features from a set of 60 features
• Created fitting curves, calculated AUC for XGBoost, random forest and GBM and selected the best model.

Here is the Github Repo

MovieBusters: Film Data Analysis

Technologies used: Pandas, NLTK, Logistic Regression, BeautifulSoup, Seaborn

• Scrapped movie data from IMDB and The Numbers websites.
• Developed a machine learning model to identify and rate actors based on the genres to maximize revenue
• Performed natural language processing (NLP) on film descriptions to filter important screenplay keywords of successful films and performed data visualization using word cloud.

Here is the Github Repo

Song Genre Detection System

Technologies used: Python, RCNN, DJango, AWS

• Built a deep learning model using CNN to identify song genre.
• Created a model to convert the song to 10 different genres
• Designed and configured the backend application using Django and hosted on AWS (Amazon Web Services) cloud.

Prediction of Mortality Rate Post Thoracic Surgery

Technologies used: Python, Logistic Regression, Random Forest, Flask, Pandas, Numpy, Seaborn, Matplotlib, SQL, Flask

• Built a web application using Flask to provide statistical backing for doctors to enhance accurate thoracic surgery predictions
• Analyzed patient history and created a Random Forest Classifier to predict life expectancy with a 98% accuracy

Here is the Github Repo for the web app
Here is the Github Repo for the pre-processing

Customer Behavior Analysis in Apps to Drive Subscriptions

Technologies used: Python, Logistic Regression, Pandas, Numpy, Seaborn, XML, MySQL

• Classified customers into likely and unlikely based on their app usages for a paid product service
• Created a model by extracting relevant features using correlation analysis and hypothesis testing to increase the customer retention rate

Here is the Github Repo

Predicting the Variety of Wine

Technologies used: Python, Seaborn, NLTK, Natural Language Processing, Tensorflow, Sequantial Model, Neural Networks

• Preprocessed user reviews using tokenization, lemmatization, stemming and vectorization.
• Implemented Latent Dirichlet Allocation (Natural Language Processing) to extract the main topics form the document.
• Implemented Sequential model to predict the variety of wine with 97% accuracy

Here is the Github Repo

QNA (Question-Answer WebApp)

Technologies used: PHP, HTML, Javascript, CSS, MySQL, AJAX, MySql, AWS

• Built a peer to peer doubt solving platform to help students resolve their queries anonymously
• Added an upvote-downvote feature to filter inappropriate posts that helped improve quality of content
• Successfully hosted the application on Amazon Web Services cloud which was used by 10000 students.

Here is the Github Repo