As a developer, I specialize in creating modular and scalable front-end responsive architectures. As a machine learning Enthusiast I create efficient high accuracy machine learning models using scikit-learn and tensorflow
My Works in the domain of machine learning include neural networks trained based on kaggle datasets, image-classification using convolutional neural networks and opencv.
In the domain of front end web development, responsive websites have been created for business clients as a part of certification projects.
Projects
Titanic EDA Prediction
This is a model on the EDA dataset on kaggle to detect whether a person, based on his/her age, ticket number, etc. would survive on Titanic. This model was created using sequential Api of tensorflow.keras module with 4 input neurons - each corresponding to a feature, 5 hidden neurons with relu activation function and one output neuron with a softmax function. The model is of 0.89 accuracy or in other words 89% accurate on the test dataset provided.
Face Detector
This is a face detector trained using Convolutional Neural Network (CNN). This model is trained using 1500 positive image samples over 200 epochs. This model is able to recognize a human face upto 90% accuracy. This model is created to build other models on top of this. This model could be thought of an abstract class. Models like mask detection, sentiment analysis can be built on top of this model by training the required samples over this model to reduce the process of training the whole batch again.
Human Motion Detection
This model detects human movement using a tensorflow model exported into opencv module. It first YOLO algorithm to detect humans, then uses relative positioning algorithm to detect movement. This model is designed to work on CCTV cameras. This could be used for burglary detection even in dark areas if cameras with infrared capabilities are used to feed input to the program. Any irregular movement may be detected by the program and the alarm will warn the authorities about the illegal entry. Motion detector sensors also have a wide application in home and personal security. It is however, impossible to have a security system without a motion sensor.
Vehicle Counter
This model detects and counts the number of vehicles passed through a particular region in a particular time. It first YOLO algorithm trained by tensorflow exported to opencv in order to detect vehicle in a given region, uses opencv to draw a boundary on to it and calculates the centroid of the given region. A hypothetical line is taken as a reference in a direction normal to vehicle movement; then the given centroid is checked if it crosses the line and the vehicle count is incremented. Vehicle motion detector has a wide application in field of traffic and safety of passengers of vehicle. It can be used to detect lane changing , turning, and speed measurement.
Website - Little Lemon Restaurant
This website has been created as a coursera project for end of course assessment for the course Advance HTML and Css by Meta. The website is created for a fictitious restaurant known as Little Lemon. The website is responsive and scales over all screen sizes.
Just another Dynamic Website
This website has been created as a part of learning Javascript.
Ain't Much, But honest work :)