Aravind Soundararajan

• Studied the data flow in the BlueStore backend of Ceph filesystem.
• Profiled and visualized the call traces during a write request on HDD and RAM disk.
• Analysed the overheads in the storage software that arise due to faster disk devices and concluded results about thread management and communication overheads.

Technologies used:

• Python
• Ceph Filesystem
• perf
• Flame Graphs
• Shell

• Incorporated CORDS, a filesystem fault injection framework to study the behaviour of Ceph in the presence of faults and corruptions.

Technologies used:

• Python
• CORDS
• Ceph Filesystem
• Shell

• Designed and implemented database indices using machine learning models that learn the data distribution from the training data.
• Evaluated the performance of learned indices on standard read-only workloads and compared it against traditional B-tree and Bloom filter indices.

Technologies used:

• Python
• TensorFlow
• scikit-learn

Consists of the following stages:

• Stage 1: Information extraction from natural text
• Stage 2: Crawling and extracting structured data from Web pages
• Stage 3: Entity matching
• Stage 4: Integrating and performing analysis

Technologies used:

• Python
• scikit-learn

Adversarial Machine Learning project to create adversarial examples for tweets. Built a sentiment analyzer based on a DNN. The DNN has one hidden layer and two dropout layers. It performed with about 80% accuracy on the set-aside dataset.

The adversaries were generated using the following two methods.

• Synonyms for words which contribute most to the label
• Insertion/Deletion/Modification of important words

Technologies used:

• Python
• TensorFlow
• scikit-learn

• Implemented a ID3-like decision-tree learner for binary classification.
• Implemented Naive Bayes and TAN incorporating Laplace estimates.
• Implemented logistic regression using stochastic gradient descent.

Technologies used:

• Python

Implemented modern OS features such as

• Multilevel feedback queue CPU scheduler.
• Null pointer dereference.
• User space memory management and stack rearrangement.
• Thread library (locks and condition variables) for concurrency and synchronization support.
• Filesystem checker and repairer.

Implemented

• An interactive command line interpreter (bash-like shell) to understand process virtualization.
• Interprocess shared memory using mutex and signals.
• Scalable web server by adding multi-threading support to a basic web server.

Technologies used:

• C
• xv6
• GNU Debugger

• Implemented a chatbot that generates sentences using trigram language model trained on 250+ essays for the English language.

Technologies used:

• Java

• In this project, we have developed a subtitle generation system which takes a video lecture as input, extracts the audio and generates subtitles also known as Closed Captions for the video lecture.
• We have used speech recognition engine to generate the subtitles from the audio and Information Extraction to generate notes (summary) and questions from the automatically generated subtitles.

Technologies used:

• Java
• CMU Sphinx Speech Engine

• An application performing conversion of human language queries to standard SQL queries.
• Performs parsing of the input query, extracts features that are used to generate SQL queries.

Technologies used:

• Java
• Stanford POS-tagger

• Implemented an automatic issue creator for exceptions in production.

Technologies used:

• Ruby on Rails
• GitHub API
• Airbrake API
• Atlassian JIRA API

• Designed database backend of a chatbot that helps the victims of floods.

Technologies used:

• Python

• Involves Computer Vision in tracking hand and moving the cursor accordingly using webcam to control the desktop.
• An alternative to mouse and touchpad.
• Funded by Centre for Technology Development and Transfer (CTDT, CEG)

Technologies used:

• C++
• OpenCV