Projects
These are few projects I did in past and lot more will be added in the future.
Automated colorimetry using machine learning
Guide: Prof. Nisheeth Srivastava
-
We have contributed to the development of a physical device that permits automated colorimetric analysis of impurities in water samples from smart-phone based photographs taken using a custom light containment device. We contributed an image processing and machine learning pipeline to this project, that ingests photographs of colorimetric patches and outputs predicted impurity levels.
Technology Used:
Python SK-learn
Image Source: https://phys.org
Fake review detection
Course: Statistical NLP(CS779A)
Guide: Prof. Ashutosh Modi
-
We performed Fake review detection using stance detection, a previously unused approach for fake review detection. For the task, we modified and pre-processed the Yelp fake review dataset to apply the stance detection models originally proposed for fake news detection and introduced network features.
-
We Implemented various models, including a Bi-LSTM model with multi-layer perceptron and a Siamese network using GloVe embedding, Bi-LSTM, Bahdanau attention mechanism and a multi-layer perceptron network.
Technology Used:
PyTorch Python
Image Source: Financial Times
Machine Language Translator
Course: Statistical NLP(CS779A)
Guide: Prof. Ashutosh Modi
-
Implemented a Machine language translator(MLT), which translates Hindi sentences into English sentences. Pre-processed text corpus by cleaning the data, tokenisation using NLTK for English and IndicNLP for Hindi.
-
Implemented pre-trained word2vec embedding with LSTM and GRU based encoder and decoder and in others allowed the model to learn word embedding specific for the task using an embedding layer with GRU and Bi-GRU with Bahdanau attention mechanism.
Technology Used:
PyTorch Python
Network Security Analysis
Course: Computer Systems Security(CS628A)
Guide: Prof. Sandeep Shukla
-
Performed Active and Passive Reconnaissance on live wired and wireless network traffic to extract valuable information about the client using Wireshark and Nmap.
-
Performed reconnaissance on the wireless network by capturing the wireless beacons using Airodump-Ng. Attacked the WEP and WPA-2 wireless network to crack its password using Airmon-Ng, Airodump-Ng, Aireplay-Ng, and Aircrack-Ng.
-
Used Wireshark to decrypt the wireless beacons using the cracked password without authenticating with the network.
Technology Used:
Nmap Wireshark Airodump-Ng Airmon-Ng Aireplay-Ng Aircrack-Ng
Image Source: wgu.edu
Optimising Matrix Multiplication
Course: Programming For Performance(CS610A)
Guide: Prof. Swarnendu Biswas
-
Optimised the matrix multiplication operation, one of the most widely used matrix operations, has a cubic time complexity that scales poorly.
-
Optimised the operation without any additional hardware by reducing cache miss during the operation using cache level optimisations and using CPU to its fullest with multi-threaded design. Further reduced the operation time by replacing the CPU with GPU for computation using CUDA.
Technology Used:
C++ POSIX Threads CUDA
Image Source: .quantamagazine.org
Stream
Android App
-
Integration of Firebase to enable an application to update in real-time.
-
Users can create temporary or permanent channels.
-
Minimalist UI for a better experience.
Technology Used:
Android Java Firebase
Clinical Outcome Prediction for Kidney Renal Cell Carcinoma using Multi-Genomic Data
-
Data Pre-processing and cleansing by Shifting data and Normalization.
-
Dimensionality reduction using Non-Negative Matrix Factorisation.
-
Training and testing a model using Bayesian Model Averaging.
Technology Used:
R N-NMF Bayesian Model
Image Source: Reddit.com
BIller
Android App
-
Communication to the server using customized REST API to communicate to the server.
-
API implemented in PHP to communicate and process queries from an android application.
-
MySQL used as a back-end database.
-
Minimalist UI for better experience.
Technology Used:
Android Java PHP MySQL JSON
Lyrics
Website
-
Four different versions each representing a different way of showing the same lyrics in different ways.
-
MySQL as a database in earlier versions.
-
The latest version uses a web crawler to extract song lyrics.
-
With integrated YouTube control to make the experience seamless.
-
Minimalist UI in the latest version for a better experience.
Technology Used:
HTML CSS PHP MySQL Javascript JQuery
Tic-Tac-Toe
Website
-
Single And Multi-Player Options.
-
The algorithm is written in JavaScript for faster decisions and no delays in computer decisions.
-
Keep records of wins/loss/tie using web storage
-
It's very fast loading on slow internet connection compared to web-based flash games available at that time.