Plant Species Classification Model

The goal is to build a model to classify weed seedlings and crops using TensorFlow.

I have modeled a machine learning pipeline that can perform data preprocessing, data augmentation, algorithm development, training, and model deployment into AWS. Also, the model is highly resistant to overfitting due to the fine-tunning of its hyperparameters and is highly efficient both on mobile and web applications.

In this neural network architecture, I have used InceptionResnetV2 as my pre-trained model and custom layers to train the model. I have also experimented using mobilenetv2, where it also performed better. When I deploy the model API on a web application, I found out that mobilenetv2 to be lightweight and fast.

Close Project

Sentiment Analysis Web App

A web app which can distinguish the sentiment of a user review and classify the review as positive or negative.

In this project, I used SageMaker to construct a complete project from end to end. It results, in a simple web app which interacts with a deployed recurrent neural network performing sentiment analysis on movie reviews.

In this application, I have designed and developed a RNN deep learning model that performs sentiment analysis on movie tweets. This model is deployed on a web application using a publicly accessible API that interacts with model endpoints. Also, in the pipeline, I have used various NLP libraries and strategies to create a dictionary of stem words from movie tweets for model training. Moreover, I have used pytorch to train the model.

Close Project

Plagiarism Detector

A plagiarism detector that can classify a text as plagiarized or not.

In this project, I have built a plagiarism detector that examines a text file and performs binary classification, to label it as plagiarized or not. I covered some ongoing research topics to build the algorithms for feature extractions, such as containment and longest common subsequence using dynamic programming. Further, the application is deployed on an AWS instance, thus can be integrated in any web or mobile applications.

Finally, after feature extraction, I feed it to the artificial neural network. In this to prevent overfitting I used drop out layers, and as it is a mutually-exclusive class problem, therefore, I used sigmoid function.

The dataset contains the corpus of plagiarized words. To understand the collection of the data and theory behind it, I referred the this research paper.

Close Project

Image Captioning Deeplearning Model

Automatically generate captions from images.

In this project, I have created a neural network architecture to automatically generate captions from images. After training the data on (MS COCO) dataset to train the model, will test the network on novel images.

Also, I have defined a LSTM decoder which sequentially look at inputs, and embed feature vector and each word. In this way it takes all features from an image

Close Project

Landmark Detection & Tracking (SLAM)

Simultaneous Localization and Mapping for a 2 dimensional world!

In this project , I have implemented SLAM(Simultaneous Localization and Mapping) for a 2-D world. The algorithm will gather robot's sensors and motion over time, and then use information about measurement and motion to re-construct.

This is an active area of research in the fields of robotics and autonomous systems. Since this localization and map-building relies on the visual sensing of landmarks, this is a computer vision problem.

Close Project

Portfolio Website

A fully responsive website!!

I developed a fully responsive web application, which is compatible with both mobile as well as laptop screens. Moreover, in this application, I have used javascript and jquery primarily to deploy my custom logic and scripts. Also, I have used many open-source javascript libraries such as bootstrap, easpiechart, owl carousel, fancybox, github card, and few others. For styling my website, I have used CSS referencing to class and Html tags. All my custom styling scripts are in style.css and custom.css.

Close Project

Facial Keypoint Detection

This project will be all about defining and training a convolutional neural network to perform facial keypoint detection, and using computer vision techniques to transform images of faces.

Moreover, I have used Haar Cascade detector to detect all faces, and transformed them into expected tensors after preprocessing, which them used as a input for the neural network.

Close Project

Task Manager Web App

A task manager web application along with nodejs and redis client!!

In this project I have developed a Task Manager using Node js and redis client. The idea behind is to develop a higly optimized application where there will be less latency, increase throughput, and ease the load off the relational or NoSQL database and application. I have used embedded javascript templating(ejs), which is a templating engine used by Node.js. Using this I have to write very minimal code for the view(front-end). Also, this rendered data to html using get and post at the client side.

This application will be able to add tasks, which will populate in a task lists,and can to remove/delete tasks. I have used array list to sore data, and push it into redis client. Furthermore, the user can also fix a business call by enter the requisite information in a form. I have implemented this functionality using a hashmap, which store the data save it into redis client.

This system design can be integrated in many web applications, and in conjuction with other nosql and relational database; thus, will provide an efficient methodology to broadcast most frequent (high demand) data.

Close Project

Sparkify data modelling with Postgres

A startup called Sparkify wants to analyze the data they've been collecting on songs and user activity on their new music streaming app. The goal is to understand what songs users are listening to. Currently, there is no easy way to query their data, which resides in a directory of JSON logs on user activity on the app, as well as a directory with JSON metadata on the songs in their app.

This application will be able to add tasks, which will populate in a task lists,and can to remove/delete tasks. I have used array list to sore data, and push it into redis client. Furthermore, the user can also fix a business call by enter the requisite information in a form. I have implemented this functionality using a hashmap, which store the data save it into redis client.

As a data engineer I have created a Postgres database schema with tables deployed on AWS EC2 instance and designed an ETL pipeline to optimize queries on song play analysis. Also, the pipeline is able to test your database and ETL pipeline by running queries.

Close Project

ETL pipeline for Sparkiy music app with Apache Cassandra

An ETL Pipeline using Apache Cassandra!!

A startup called Sparkify wants to analyze the data they've been collecting on songs and user activity on their new music streaming app. The goal is to understand what songs users are listening to. Currently, there is no easy way to query their data, which resides in a directory of CSV logs on user activity on the app.

In this project, I have applied data modeling with Apache Cassandra and complete an ETL pipeline using Python. I have modelled the data by creating tables in Apache Cassandra to run queries. The ETL pipeline will be able to transfers data from a set of CSV files within a directory to create a streamlined CSV file to model and insert data into Apache Cassandra tables.

Close Project

Brain Tumor Detection

To Detect and Classify Brain Tumor using deep learning!!

In this project we are segmenting organs cells in images. The training annotations are provided as RLE-encoded masks, and the images are in 16-bit grayscale PNG format. The goal is to Detect and Classify Brain Tumor using, CNN or ANN; as an asset of Deep Learning and to examine the tumor position(segmentation). In this project, I will be using U-net to perform image segmentation, which has tremendous application in medical imaging. Also, finally we will be able to generalize to both partially and wholly unseen cases.

After Validating it with the test data, I found out U-net is highly efficient in locating the area of abnormality with a high segmentation accuracy. The model has delivered 99% IOU and 63.28% of dice coefficient on test data. It can be further improved by increasing the epochs.

Close Project

Data Lake with Apache Spark

The purpose of this project is to build an ETL pipeline that will be able to extract song and log data from an S3 bucket, process the data using Spark and load the data back into s3 as a set of dimensional tables in spark parquet files. This helps analysts to continue finding insights on what their users are listening to.

The data gets that gets extracted will need to be transformed to to fit the data model in the target destination tables. For instance the source data(star schema) for timestamp is in unix format and that will need to be converted to timestamp from which the year, month, day, hour values etc can be extracted which will fit in the relevant target time and songplays table columns. The script will also need to cater for duplicates, ensuring that they aren't part of the final data that is loaded in the tables. .

Close Project

COVID19 Universal Data pipeline

A universal data pipeline, that can parse various data formats and load data into single repository!!

The primary goal of the project is to define a universal data pipeline that will collect data from all scattered datasets found over various websites in different formats. Currently, various organizations are collecting data, but there is no unified platform that collects data from all sources and maintained a single repository on which researchers can work.

As the final end-user would the researchers or biotech companies, thus this pipeline will collect data automatedly and update the repository at a specific time. Moreover, this pipeline would also perform data cleaning, and transformation, therefore, will help Machine learning developers and Web Developers to use it for various applications.

I have Conceptualized and developed a data pipeline that can parse various data formats and dynamically load data into a cloud Datawarehouse. Also, the pipeline is deployed using Apache Airflow for better automation, parallel processing, seamless data scheduling and monitoring.

Close Project

Deploy a high-availability web app using Cloud Formation

A web application is deployed using cloudformation!!

In this case scenario, a company has created an Instagram clone called Udagram. Developers pushed the latest version of their code in a zip file located in a public S3 Bucket. Here, I am deploying the application, along with the necessary supporting software into its matching infrastructure. It is accomplished in an automated fashion so that the infrastructure can be discarded as soon as the testing team finishes their tests and gathers their results.

In conclusion, a web application (Apache Web Server) and pick up code (Javascript and HTML) from S3 storage and deployed in the appropriate folder on the web server. Two parts: 1. Develop a diagram that aids to understand the CloudFormation script. 2. Interpret the instructions as well as your own diagram and create a matching CloudFormation script.

Cloud formation script will create VPC, public and private subnets, EC2 instances, security groups, AZ's, routing table and rules, S3 roles and responsibilities, S3 buckets. Also, to handle traffic I attached an application load balancer to private subnets.

Close Project

Song Play Analysis with S3 and Redshift

In this project, we try to help one music streaming startup, Sparkify, to move their user base and song database processes on to the cloud. Specifically, I build an ETL pipeline that extracts their data from AWS S3 (data storage), stages tables on AWS Redshift (data warehouse with columnar storage), and execute SQL statements that create the analytics tables from these staging tables.

Datasets used in this project are provided in two public S3 buckets. One bucket contains info about songs and artists, the second bucket has info concerning actions done by users (which song are listening, etc.. ). The objects contained in both buckets are JSON files. The Redshift service is where data will be ingested and transformed, in fact though COPY command we will access to the JSON files inside the buckets and copy their content on our staging tables.

Moreover, in this project all the resources which include S3 bucket, custom roles and responsibilities, Redshift cluster on an EC2 instance are being deployed using python SDK boto3 library.

Close Project