Hello!

A simple tool designed to help navigate the diverse range of government schemes in India, catering to sectors like startups, education, agriculture, and more. The goal? To make finding relevant information more accessible and more efficient for everyone.

A react based scalable messaging application built with industries best practices and currently leading technologies in the ecosystem, with modularity and performance as top metric.

An IOS Application using CoreML and ARkit for 3-D model transformations by recognizing hand gestures in realtime.

Duplicate question detection using Ensemble model of Siamese Network and XGBoost with Glove Embeddings. Our submission was among top 10% and later model was improved with state-of-the-art BERT embeddings to achieve accuracy of 92%.

The goal of this project was to implement and train a named-entity recognizer (NER). Most of the feature builder functionality was implemented using spaCy, an industrial-strength, open-source NLP library. For classification a maximum entropy (MaxEnt) classifier is used. Different contextual, meta features and machine learning techniques such as google word embeddings and clustering is used to increase the performance of the classifier.

A reinforcement learning Automated tutoring agent that interacts with the student and choose the activities/topics that help students maximize knowledge by selecting topics to review sequentially and effectively, which is also known as spaced repetition learning . Spaced Repetition is formulated as a POMDP environmentin the paper and our primary contribution is to create an automated tutor which takes an observation of student answering question and chooses the next question to review or learn.

Big data project to analyze safety of cyclist in densely populated cities such as NYC. With the increasing popularity of the bike sharing programs in NYC, there has been increasing concerns regarding the safety of riders, mainly because the NYC has high traffic and also it is not mandatory for bike sharing riders to wear helmets. This paper performs the big data analysis on NYC Vision Zero, Citi Bike, and Weather dataset to analyze the pattern between the accidents and city bikes. We use various big data technologies such as MapReduce to perform data cleaning and ETL, and then analyzed data using Hive and HBase.

Analyzed the trending youtube videos to show how the cultural divergence affects people’s likes and analyze most popular video types to shed light on how YouTubers can refine their videos to get more subscribers, and also recommending popular channels for different video genres.

An augement reality application which can be used to teach young students about our solar system. Students can interact, change the speed, gravitation, rotation of planets and learn various information about them. Provides an interactive and fun way of learning.

Built an fault tolerant MapReduce based key/value system which consists of master server and worker nodes. Raft Consesus algorithm was implemented from scratch that maintains replication, strong consistency and garbage collection of old log entries.

Designed a fault tolerant, reconfigurable shard key/value storage system in Golang for increasing throughput performance using replica groups consisting of servers that uses RAFT consensus.

Presented detailed analysis of the effects of parameters on the CNN and SVM models for image classification and proposed an Ensemble-Hybrid model which obtained classification accuracy of 94.6% on CIFAR-10 dataset and optimized the hyper parameters using adaptive genetic algorithms.

Implemented various state of the art RL algorithm such as PPO, A2C, A3C using PyTorch and OpenAI which will be used for tutorial on my medium blog. [GitHub]

Designed mobile app for dementia patients with features such as tracking medication and medical appointments, identifying nearby pharmacies and service providers, emergency button and offering cognitive learning games.

Simple chat application using Node js, react, express. It was built for the final project of my undergraduate web dev course, and I think I nailed it as I got an A in the course. But most of the credit goes to my Medium clone as by doing that project, this one was piece of cake for me.

Medium Clone web app built using MERN Stack. It was built for fun as I was interested in learning React.js and grow my skill set. So it took lot of weekend hours, and I loved the result.