GuDB - fork of Apache Spark Catalyzer

Feb - May, 2021

  • Gutted version of Spark Catalyzer provided - had to build a Database that works efficiently for SELECT queries
  • TPC-H Standard Queries were used to benchmark the database on 0.1 Scaling Factor
  • Pure Scala implementation with leveraging Spark Catalyzers Logical Plan, types, APIs
  • Implemented View Materializer, storing Statistics on Lineitem Tables to achieve faster performance for Query 1
  • Query Optimizer traits developed using Relational Algebra Tricks
  • Building Indexes on tables - to implement SortMerge Join and faster equality joins
  • Course Project for CSE562, implemented across semester with 4 Checkpoints

Mentors

Prof. Oliver Kennedy

Software Defined Routing

Apr - May, 2021

  • SDN - Graded Controller, with N Routers defined with an initial topology, where the network will use Bellman-Ford algorithm to stabilize the graph
  • Pure C++ with inet, socket programming with every Router implementing context switch based on the select() API in sockets
  • Implemented Crash, Weighted Updated, UDP Gossip among Neighbour Routers
  • Data Plane of forwarding packets from source -> destination based on shortest path
  • Course Project for CSE589

Mentors

Prof. XiangHang Mi

RL Hyperparameter Tuning with Bayesian Optimization

Oct - Dec, 2020

  • Tuning DQN, Q-Learning RL hyperparameters using Bayesian Optimization for OpenGym environments - Taxi (Grid), Lunar Lander(Box 2D)
  • Influencing Explore/Exploit trade-off for standard Bayesian opt techniques using Gaussian Parameters (Kernel, Mean) and formulae parameters of EI and others
  • Comparison with other standard optimzations, and studying the various Optimzation Techniques
  • Worked on Taxi Problem, analyzing all algorithms EI, PI & Thompson Sampling influencing the optmization by controlling gaussian and algortihm parameters
  • Course Project for CSE610

Mentors

Prof. Varun Chandola

Team

Classroom Scanner

Sept - Dec, 2020

  • Classroom Scanner, a solution for the K12 community teachers which help in uploading of Classwork directly to coded folders scanning from the application
  • Teachers create a Coded Folder in the Folders UI, students write the code on assignments which on scan gets submitted to the folder
  • On Scan of documents, the system has an OCR snipper to extract the header section of top page and parses to find the Assignment code to reduce the manual work of teachers in archiving these documents
  • Project developed with ReactJS, Laravel & React Native (both iOS & Android)
  • Developed the Frontend interface for the project including the Mobile application

Mentors

Prof. Alan Hunt

Prof. Jeff Brunskill

Team

Dijkstra's Algorithm Verification with Frama-C

Nov - Dec, 2020

  • Dijkstra's algorithm, is an example of a Greedy Alogrithm to find the shortest path of a graph with no negative weights
  • We used Frama-C to prove the properties in the algortihm to prove the correctness - Relaxation Property, Path from Source
  • Every reachable vertice from Source, has a Stack which indicates the path from the source
  • Stack data structure operations also were proved using Frama-C (non-mutation & mutation properties)
  • CSE715 Seminar Project

Mentors

Prof. Bharath Jayaraman

Team

ML Fairness in Judicial Systems

Apr - May, 2020

  • Designing a new System for the US courts, taking ProPublica's dataset on analysis of COMPAS system
  • Using various Fairness Techniques and unbiased features with no direct co-relation to either race/sex
  • tasked to design a system with Feature Engineering, Model selection and using Fairness techniques of post-processing
  • Including Secondary optmization criterion to achieve a lower cost, as incorrect classification cost the Government and the public
  • CS574 Internal - Market Submission Winners

Mentors

Prof. Varun Chandola

Team

Pintos OS Development

Feb - May, 2020

  • Pintos C based operating System emulated on Bochs, task was to develop the incomplete modules of the OS
  • Project 1 - Kernel Level Tasks - Synchronization, Alarm Clock, Priority Scheduler, Multilevel Feedback Queue Scheduler
  • Project 2 - User Level Tasks - System Call Infrastructure, Secured RW for executables, Argument Passing for Programs, Process synchronization

Mentors

Prof. Farshad Ghanei

Team

Movie Genre Prediction with PySpark

Apr - May, 2020

  • Movie Synopsis - Parsing and Processing, using various Feature modelling to represent Data like CountVec, Word2Vec, IDF ...
  • Using various ML classification algorithms for best accuracy
  • Kaggle Internal Competetion - Placed 2nd/90 teams in Public Leaderboard

Mentors

Prof. Deen Dayal Mohan

Team

FinChain | Invoice Automation

May, 2018

  • Problem Statement: Design a end-to-end Tracking System for all parties invloved in Supply-Chain of invoice
  • Prototype developed Python Flask, ReactJS & MongoDB for tracking invoices and email alerting system for concerned parties on every step of progression
  • Experimented with Hyperledger Composer to build a similar system of invoice tracking and alert automation
  • IBM Business Hackathon 2018 - Intra IBM Competition - Winners

Mentors

Supriya, IBM Sr. Manager

Jayabharath, IBM Sr. Developer

Team

Hawkeye

Feb - Apr, 2017

  • An image based object recognition for reconnaissance purposes from Aerial vehicles.
  • The project was built with OpenCV, Python, Keras w/Theano
  • Embedded Solution, with image foreground extraction done on RaspberryPi and forwarded to server
  • Python Flask server proccesses incoming image and stores the identified object and timestamp
  • Image Classification done using Deep Learning with the pre-trained VGG16 model
  • Developed at Smart India Hackathon 2017 Finale.

Mentors

Prof. Bhaskar

Prof. Subhatra

Team

Braillity (Braille Reality)

Feb - Jun, 2016

  • Solution for the visullay challenged, where a person wears an embedded device with a camera, which processes images and feed to a braille pad
  • The project was built with OpenCV, Python
  • The solution was to indentify object and process the outline of objects and feed it to a braille pad
  • The prototype developed was able to process shapes, outlines of objects but the Braille pad was only a visualization
  • Developed at Anna University, Titan Hackathon 2016 - Finalists

Mentors

Prof. Bhaskar

Team