Projects

Yelp Recommendation Engine

Mar 2020 - May 2020

Designed Recommendation engine in Python(Hadoop, Pyspark and SparkSQL), that predicts the User rating for any Business(on a scale of 5), and displays the top-rated Businesses, by implementing concepts of User-Based and Item-Based Collaborative Filtering.
Scaled the engine (by tuning hyper-parameters) to work within 300s on Yelp Training Dataset of size over 1.5GB.
Implemented a Hybrid Recommendation model, with Pearson Co-relation and Feature extraction using XGBoost, and removed Cold Starts, to obtain a Root Mean Squared Error under 1.0, and accuracy of 98.7% against the Testing dataset.

May 2020 - Jul 2020

Created a Task Manager App using NodeJS, MongoDB, Postman(for serving HTTP requests), and Heroku(for Deployment).
Features of the app include New User Registration, User authentication, adding and modifying tasks, image upload, GPS location services using MapBox API, and deadline reminder notifications on User Homepage.
Incorporated unit & integration tests to ensure smooth functioning of CRUD operations, and HTPP requests.

Oct 2019 - Dec 2019

Built a social media graph in Pyspark, by connecting Users that have rated similar businesses(by adjusting the similarity threshold).
Implemented the Girvan-Neuman algorithm in Python/SparkSQL, to find the betweenness values for 1500 edges.
Used these betweenness values to perform hierarchical clustering of the graph into tightly-knit communities, by maximising a Modularity function.

Jan 2020 - Feb 2020

Created a Jupiter Notebook to classify a set of 5574 SMS texts into SPAM or HAM, using concepts of Text Mining(TF-IDF), and Naive Bayes.
Performed data preprocessing steps, including removal of punctuations and stop words, and converting cleaned data into Count Vectoriser format.
Then performed Multinomial Naive Bayes Classification, to separate messages into SPAM or HAM with 98% accuracy.

Jul 2020 - Aug 2020

Created a budgeting using React, Hooks, Redux, and Firebase.
Features of the app include user authentication via Google Firebase; add, modify, delete expenses; and display expenses sorted by day, month, or year.
Added support for images, geo-tagging, and expense charts in the UI.
Deployed the app using Heroku.

Sep 2020 - Nov 2020

Crawled nytimes.com and used Apache Tika to parse and load the crawled files into an Apache Solr core.
Extracted the link structure of the crawled files using Java's JSoup library, and computed PageRank for each page, using Python's NetworkX library.
Created a PHP web application to compare the search results for Lucene text search vs PageRank.
Used Ajax and JQuery to implement auto-complete, snippets, and spelling correction functionalities for the search queries..