Projects
NFT Marketplace
Created a service-oriented REST architecture based marketplace web app with CI/CD, load testing and dashboarding. Implemented the Infrastructure-as-a-Code (IaaC) paradigm and scaled the application to 5000 concurrent users and 2800 requests/second using simulated load testing via Gatling.
Crime Hotspot Prediction
Experimented to utilise a novel Convolutional LSTM model to extract spatio-temporal features to predict crime hotspots over Vancouver city with 82% recall and 71% precision. Worked on developing a data collection api to extend model training for multiple cities and predit crime-wise hotspots.
Aviation Safety in Canada
Collaborated in a team to use Spark, AWS Lambda, S3 and Glue to implement an ETL pipeline to scrape and process civil aviation accident and incident records on a weekly basis and AWS Quicksight for dashboarding. Developed an unsupervised system using K-Means clustering algorithm to determine the possible features leading to fatalities/injuries in an accident.
Advanced Regression techniques on House pricing datasets
Performed EDA on the Advanced Regression: House pricing dataset on Kaggle. Examined various properties like normality, skewness, kurtosis as well as performed log transformations and analyzed the differences using a normal probability plot.
Emof_AI
Emof_ai is a web API which takes a sample video as an input, detects all the faces in each frame of the video and performs classfication. It classifies the emotion of the face into 1 of 8 emotions viz anger, disgust, fear, happy, sad, surprise, neutral and none(this is when there are no faces detected in the frame at all and hence none means no emotion detected in the frame at all). It also classifies the gender of the person. In addition to that, it also classifies the age group of the person into 1 of 4 age groups viz adult, child, old and youth.
Event Analysis using Spark and Elasticsearch
This project aims at building a scalable data analysis pipeline that can ingest events using the New York City leading causes of death data and process them using tools like Apache Spark and Elasticsearch.
Fraud Detection
Performed Exploratory Analysis on fraud detection dataset and tried to find interaction terms. Moreover, compared various classifiers offered by the scikit-learn library
Molecular Classification of Cancer by Gene Expression Monitoring
Analysis of the 20 NewsGroup Dataset followed by lemmatization and k-means clustering and non-negative matrix factorization for topic modelling
Topic Modeling
Analysis of the 20 NewsGroup Dataset followed by lemmatization and k-means clustering and non-negative matrix factorization for topic modelling
Blogging website
Created a blogging website using Flask framework as well as used SQLAlchemy for building relationships and CRUD operations. Used Elasticsearch for performing searches. Deployed on Heroku using PostgresDB and Searchbox ES add-ons.
Sentiment Analyzer
Created a service based on a Tensorflow model which employs Convolutional Neural Networks to perform sentiment analysis to do binary classification. Used TfServing and Docker to orchestrate modeling as well as inference pipeline