MedBuddy
A web platform which connects doctors, patients and services like ambulance provider. Patient can choose a date and doctor to consult, the system will automatically generate appointment time based on doctor's schedule. Patient can also view his current and past appointments and book ambulance service. Doctor on the other hand can accept/reject the appointment, view all his current and past appintment and upload detailed prescription for the patient. The website is hosted at https://medicalbuddy.herokuapp.com/
VITON - a Virtual try on system
TryFirst is an interactive web-app which benefits users by allowing them to try-on a particular cloth to see how it suits them. Thus, allowing consumers to virtually try on clothes will not only enhance their shopping experience, transforming the way people shop for clothes, but also save cost for retailer.
Bird Call Identification
Identifying a wide variety of bird vocalizations in soundscape recordings using CNN by generating melspectrogram images of audio. There might be anthropogenic sounds (e.g., airplane overflights) or other bird and non-bird (e.g., chipmunk) calls in the background, with a particular labeled bird species in the foreground which we are finding using our machine learning models
Weed Whacker
Deep-learning model to identify and classify weed plants deployed in an android application. By simply pointing your phone at the plant and a single snap, the app will tell you whether the query plant is a weed or not along with its full description.
MNIST Regeneration using Generative Adverserial Network(GAN)
Recreated the Handwritten Digits from the MNIST Dataset using a Generative Adversarial Network(GAN), an implementation of "Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks" (https://arxiv.org/pdf/1511.06434) paper.
B+ Tree Implementation in Hadoop For Big Data
Implemented the indexing of B+ Tree on Hadoop Distributed File System on top of HBase database to make efficient queries such as: Search by primary key. Time Complexity: BigO(log(n)/(number of clusters) Search by a non-primary key. Time Complexity: BigO(n/(number of cluster)) Insertion and deletion of a single row in the database. Time Complexity: BigO(log(n)) The indexing is very effective in the scenarios where huge datasets can?�t be loaded on RAM so in those cases this method of indexing load just a single node of the dataset at a time. Also, as the size of a single node is large, it takes fewer memory checks to perform queries.
WalkSafe
A smart system for pedestrians wearing earphones, that alerts the user whenever he/she is being honked at by an oncoming car. Promises to dramatically reduce the no. of accidents caused due to pedestrians crossing streets wearing earphones. Starting 26-10-2018.
Advanced sentiment analysis based Twitter support bot for #MeToo
Used Natural Language processing and Machine learning algorithm to do a Sentiment Analysis and grade a #MeToo post. Validated the genuinity of the post on the basis of polarity and factuality. Analyzed emotion of the post & performed tasks such as re-tweeting the genuine post, and commenting supportive solution.
Recommendation Engines
Implemented different Recommendation Engine models on Collaborative filtering and Association based algorithms. 1 - Collaborative Filtering based on Nearest Neighbors using Hamming distance. 2 - Collaborative Filtering based on Nearest Neighbors using Pearson's correlation. 3 - Collaborative Filtering based on Latent Factor (It solves sparse matrix problem and is highly efficient too). 4 - Collaborative Filtering based on Regression and uses Gradient descent algorithm to reduce cost. 5 - Association based Recommendation Engine based on Apriori Association.
