Developed a complete Multilingual search engine which retrieves Tweets based upon queries, with focus on ingesting, indexing and searching twitter data in 5 different languages. The retrieved results were shown along with summarization and faceted search filters to refine search. The data was indexed using IR models - VSM, BM25, Language models and were evaluated on various metrics like precision, recall, ndgc, bpref values. Using SolrLucene and Java, experimented with various retrieval models, tagged the content with named entities like Wiki links.
Parallelizing data processing using Hadoop MR. Design of interesting analysis questions and solutions (Experiments) of University room Allocation and Course and Student Enrollment Data over the years. Building an analysis and visualization user interface using Tableau
Implemented a multilayer perceptron neural network in Python and evaluated its performance in classifying handwritten digits from the MNIST database. Achieved maximum test accuracy of 94.6% in 60 sec.
Implemented a Dynamo-style key-value storage.There are three main features: 1) Partitioning, 2) Replication, and 3) Failure handling. The project provides both availability and linearizability at the same time. In other words, the implementation always performs read and write operations successfully even under failures. At the same time, a read operation should always return the most recent value.
DHT based on Chord. Three features implemented are: 1) ID space partitioning/re-partitioning, 2) Ring-based routing, and 3) Node joins. The project implements all DHT functionalities and support insert and query operations. It is capable of running multiple instances of the app, all content provider instances form a Chord ring and serve insert/query requests in a distributed fashion according to the Chord protocol.
Gradient Domain fusion – developed using MATLAB, the project had features like Poisson blending and mixed blending and was used to make a single blended image from two images. Developed a user friendly UI and efficient fast generation of the blended images for easy save.
Please get in touch - I'll get back to you as soon as I can.