- Leading a team of students and researchers to apply statistical methods to imbue accountability to the sharing of digital content, audit social media governance policies, and model how information spreads online. - Applying machine learning to help local and global newsrooms, fact-checkers, non-profits, and civil society organizations to better understand their impact and audiences. - Working with The Times & The Sunday Times, Deutsche Welle, the Yale Daily News, Fundamedios, and others. Check out: https://parrot.report

- Outlining metrics to evaluate model performance at different stages over time to understand the model's contribution to the business and to perform new data source impact assessment.- Updating ML process, including researching solutions, updating feature generation process, generating labeled data, generating prototype ML models, updating unit tests, and deploying changes to production- Researching NLP solutions to enhance B2B intent signals, expand the customer base, and improve customer onboarding.Quantified false association of single IP to multiple domains, created labeled dataset, trained and evaluated the prototype IP2D model to better detect and remove these associations.

- Examined and visualized the temporal evolution of the B2B data distribution using kernel density estimation, Kolmogorov–Smirnov test, and geographic distribution of IP classes using geospatial visualizations. - Performed time series analysis and hypothesis testing to study the effect of COVID-19 on ip-type-features- Implemented dimensionality reduction methods like PrincipalComponent Analysis (PCA), Linear Discriminant Analysis (LDA), t-distributed stochastic neighbor (t-SNE) and clustering algorithms K-Means clustering, Hierarchical clustering, DBSCAN, and Gaussian Mixture Models to have the same class data to cluster together and be as differentiable from other classes as possible and analyzed their performance.- Implemented Logistic Regression, Random Forest, and XGBoost classifiers for IP-type classification into different domains and analyzed their performance for different data splitting strategies using Precision, Recall, F1- Score, and ROC-AUC and prediction probability density curves as evaluation metrics.

- Developed an NLP tool - Gumshoe to help journalists identify task-relevant text in large email corpora and evaluated the performance of the tool for multiple datasets.- Documented findings while highlighting their limitations and potential for bad or potentially misleading results and communicate results to the multidisciplinary team.

- Scraping and querying public funding/grantmaking databases, inclusive of those at the United States Dept. of Agriculture to create simple databases and high-quality visualizations.- Named Entity Recognition and Topic Modelling to ascertain how public dollars were spent on a particular grant and project.

Worked under the mentorship of Prof. Mohamed Zahran from NYU Courant to predict the performance of a parallel program on a parallel machine based on performances of previous program on other machines.

Worked under the Prof. David Kanter from School Environment NYU to Analyze and visualization of trade-offs within the new Sustainable Agriculture for the time series data available for 216 countries from 1960 to 2016.

- Analyzed and preprocessed signals generated by ultrahigh-energy cosmic rays (UHECRs).- Designed a neural network to predict Xmax value for the UHECRs using time series signals and stationery information about detectors to understand the origin of these rays with the error of 0.12.

- Processed million-row data from CRSP and CompuStat, merged various tables through SQL to get daily stock prices and returns for S&P 500 index constituents.- Implemented Regression models to model the impact of news signals- derived using NLP techniques like Named Entity Recognition and Entity based sentiment score on market volatility.

We collected a data using motion sensor which was placed in the pockets of the user on a timely basis for subjects aged between 18- 48 years. Data is collected for total activities like - sitting, sleeping, standing, walking, walking upstairs, walking downstairs, running, etc.. For each of the movement 3 values is recorded namely- gravity, total acceleration, and gyroscopic value. These feature values are further normalized and scaled to avoid ambiguity in the data and to produce more accurate results. RNN-LSTM is used for the classification of data to identify different activities on the basis of gender, age, weight, and height.This information is then used for the mining behavioral patterns and provides better patient recovery training guidance to manage and reduce the risk of many diseases such as obesity, cardiovascular, and diabetes.

During the course of the Internship, I learned to develop technical specifications and plans to meet the user's requirements. I was involved in producing documentation and necessary charts and diagrams presenting the flow of the modules and developing software according to the client's needs, debug, and deployed them for use in the market.