August 2016

It is a system to detect depression in subjects from a dyadic interview sessions(DAIC WoZ dataset). Designed and modeled a multimodal machine learning architecture to combine acoustic (voice frequency, loudness, tenseness etc.), visual (face expressions, eye gaze etc.) and linguistic (word level features like pronouns, word emotions etc.) features, in order to achieve the final classification. Our system also estimates the severity of depression using multimodal regression techniques. Language/Technologies used: Python (Numpy, Pandas, sklearn, NLTK)

August 2016

GeoShare is a Social Networking application with geo-spatial features. A user can follow other people. A user can post geo tagged status. Once his friends enter the area visited and checked in by him, they will be able to see what he posted while he was at that place. Users can also view the posts through Vuphoria where the posts are augmented to the real world screen coordinates.

March - May 2016

Created a system that can summarize news articles (Spanish in particular), using keyword extraction techniques. Our approach Improved the TextRank algorithm to generate the summary of a body of text with an average ROUGE-N Score of 0.37, a 69.92% increase over the original TextRank summarizer. TextRank 2.0 incorporates a sentence ranker that scores the sentences in a text body based upon their positional probability.

April 2016

Implemented a Hidden Markov Model (HMM) Parts Of Speech tagger. The Penn Treebank was used as the training data. The sequence was found using the famous Viterbi Algorithm in python

September 2015

Designed a game agent that comes up with the best possible next move for a player by implementing different AI algorithms like greedy-best first search, minimax algorithm and alpha beta pruning which works by constructing a game tree.

December 2015

Developed an engine to determine if a query can be inferred from the given knowledge base using backward chaining algorithm in First Order Logic.

Color and Object Detection in an image using Map Reduce and PyCUDA

June 2015

The color and objects in an image were detected using Map Reduce technique on the CUDA platform to achieve better efficiency. The pixel values of an image are obtained in the "Map" stage while they are grouped using K means clustering in the "Reduce" stage. For the python wrapper for CUDA, PyCUDA was used.

July 2013

An automation system that lets differently abled to operate the home appliances like lights, fans, televisions using wireless devices. Our System can process the user’s voice input sent through the Android application and send it to the Intel Galileo. The system can also be used for personalizing as well as automatizing according to the user's wish. (The MAC address of any wifi enabled device has to be pre registered). HoPePi recognizes the device as soon as it comes into the vicinity of the wifi range and act accordingly. The Connected devices are reported to Cloud Server and the App user can also see which member is in house.

Go Mad

March 2014

It is an Android application that plays two songs simultaneously in a single earphone, one song in the left ear plug and another in the right. It is one of its kind that lets two people share a single ear phone without getting compromised on the playlist.

University of Southern California

Foundations of Artificial Intelligence (CSCI 561)

Database Systems (CSCI 585)

Analysis of Algorithms (CSCI 570)

Applied Natural Language Processing (CSCI 544)

Multi-modal Probabilistic learning of human communication (CSCI 535)

Geospatial Information Management (CSCI 587)

M.S.Ramaiah Institute of Technology

Object Oriented Programming

Web Technologies

Computer Networks

Data Structures and Algorithms

Advanced Programming with Java

Database Management Systems

Embedded Systems