Pallavraj Sahoo

New York City · New York · (412) 327-1298 ·

Data Scientist with 3+ years of professional experience in Data Analytics, Software Engineering, and Machine Leanrning and a master's degree specialized in Data Science from Carnegie Mellon University.

Industry Experience: Health care (CitiusTech), Entertainment (Sony Pictures Entertainment), and Manufacturing/IT (Caterpillar Inc.)

Skills

Data Analytics 90%

Machine Learning 90%

Deep Learning 85%

Tableau 85%

Python 90%

Java 90%

SQL 90%

Microsoft Azure 80%

Projects

Amazon Review Rating Prediction

Carnegie Mellon University

Performed Text data processing and EDA on review texts to remove noise and handle imbalanced data and outliers
Constructed numerical features for prediction by implementing frequency-based features like TF and TF-IDF
Implemented, trained and evaluated various Machine Learning algorithms like Logistic regression, SVM and Random Forest.
Achieved accuracy of ~82% and deployed the trained model to a public API endpoint on Azure ML Studio

Code

Image Classification using Convolutional Neural Network and Logistic Regression

Carnegie Mellon University

Built and trained Logistic Regression and Convolutional neural network (LeNet, AlexNet) models based on CIFAR-10 datasets
Achieved an accuracy of 85% by comparing performance of different models using PyTorch
Created a public endpoint API and deployed model on Azure ML Studio

Code

Question-Answering System with SQuAD dataset

Carnegie Mellon University

Used NLTK’s parse trees and SpaCy’s semantic rules generated question from a Wikipedia article
On a pretrained GLOVE model, Generated Answers by resolving co-references and measuring similarity between Q&A
Using probabilistic context free grammars (with Viterbi algo), to evaluate Grammar correctness of a sentence

Code

Recurrent Neural Network based Sentiment Classification

Carnegie Mellon University

Used pre-trained BERT embedding to train a sentiment classifier on Twitter tweets
Built a custom-defined loss function on PyTorch to evaluate the model performance
Achieved ~20% increase in accuracy from baseline, by optimizing the training and tuning hyper-parameters

Code

Social Media Influence on Crypto Space

Carnegie Mellon University

Performed Topic Modelling (LDA) on scraped tweets of Crypto influencers to find their influencer clusters
Assessed the inter-cluster variation by performing sentiment analysis and mapping correlations to original tweets
Performed sentiment analysis to find impact of Bitcoin on the prices of other cryptocurrencies

Code

Optimizing Product Stocks and Orders in an E-Commerce

Carnegie Mellon University

Performed Data Pre-processing using Jupyter Notebook in Python and Exploratory Data Analysis using Tableau
Provided Reorder Prediction and Recommendation using Decision Trees
Implemented Market Basket Analysis with Association Rule Mining using SAS Enterprise Miner

Code

Certifications