Aparna Iyer

Intro

• GitHub
• LinkedIn

Hello, I am Aparna, studying Masters in Computer Science at the University of Illinois Urbana-Champaign. I worked as a Software Engineer for ConnectWise, India. I intend to use this space of mine on the internet to write about my professional experience and hobbies. I graduated with a bachelor’s degree in Computer Engineering from Mumbai University. Over my two years working in the corporate, I have written automation scripts in Karate to expedite testing for business management and security products. Also, I have actively contributed to improving the automation framework and detecting critical vulnerabilities while working with the Quality assurance team. Further, I have dabbled around data science and analytics and have worked on several data analytics projects before. If not for coding, you’d often find me playing the guitar and reading non-fictional books. Feel free to reach out to me by clicking on the above links if you think I could be of any help.

Skills

Python, Git, Project management, Data Structures and Algorithms, Data Analysis, Microservices, Automation, Quality Assurance

Projects

• Project 1: Tele-dermatology application

  Skin conditions such as acne, rosacea, and pigmentation may not be life-threatening but can disturb the person experiencing it. For such benign skin conditions, people often avoid going to the dermatologist either due to monetary issues or lack of time. But if left untreated, these conditions could aggravate in time. The project aims at developing an android application that can be accessed anytime and anywhere for instant diagnosis and indicating the probability of skin abnormalities like rosacea, pigmentation, and acne. Once the anomaly has been detected, the application aims at recommending products for treating the same. It also shows nearby hospitals where the dermatology department is present on Google maps. Used Tensorflow Lite to deploy the app on the android device. The images were trained using a transfer learning approach, and MobileNetV2 was used. The system detected the abnormalities viz: rosacea, pigmentation, and acne with an accuracy of 80% on the testing set. Tech stack used: Android Studio, Tensorflow Lite, CNN, Tensorflow, Keras, MobileNetV2 for transfer learning.

• Project 2: Garbage Profiling System

  About 2 billion tons of garbage is generated annually, and one-third of it is mishandled. A model developed using Mask R-CNN to aid the garbage profiling process was designed. The dataset used for this study was obtained from Kaggle and a few external sources. After cleaning and labeling the data, the model was trained using the Mask R-CNN algorithm to classify plastic waste. It was deployed on a website and visualized the results using Python's Pygal library to make the model user-friendly.

• Project 3: Exploratory Data Analysis

  Analyzed the US accidents dataset using Numpy, Python, and Pandas.

• Project 4: Potato Disease Identification

  Potato harvest is affected mainly because of two common diseases- early blight and late blight, and because of this, farmers growing potatoes face a lot of economic loss. This project aims to detect these diseases early to prevent financial loss by accurately identifying the type of disease. Data cleaning and pre-processing were done using the tf dataset and data augmentation. The model was built using CNN for image classification and then exported. For the mobile application part, this exported model was converted into a tflite model using quantization and then deployed on google cloud to make the classification process more user-friendly.

• Project 5: Scrapping GithHub using Beautiful Soup

  Scrapped the website - https://github.com/topics. To get a list of topics. The top 25 repositories were selected for each topic, following which topic title, stars, topic page URL, and topic description were fetched and populated in a CSV file. Requests library was used to download the webpages, used Beautiful Soup to parse and extract information, and used Pandas data frame to create the CSV file.

• Project 6: Automation using selenium

  Automated the Selenium easy website using Selenium

• Project 7: Instagram automation using Python

  Automated the actions- uploading a picture, following, unfollowing, sending direct messages using Python's InstaBot library

• Project 8: Smart Medical Report generator

  During an accident or mishap, the victim may not be conscious, making it difficult for doctors to know about their previous medical history. This system makes it easy to access the medical record of the victim. An RFID tag is assigned to each person. This tag will hold the previous medical history of the patient, which can be updated regularly by the doctors. Only selected doctors with stored records can view the patient's medical history. This system saves time and money, which the patient would have invested if they started treating existing ailments from scratch.

Papers published

• Paper 1: A Skin Disease Detection System Using CNN Deep Learning Algorithm

  Springer Nature, Singapore, International Conference on Soft Computing and Signal Processing

  In this paper, a model developed using deep learning has been presented. This model distinguishes healthy skin from the skin with abnormalities such as acne, rosacea, and pigmentation and suggests products for its treatment. A repertoire of 5000 images was made for this study through Dermnet.com and other external sources. Transfer learning has been used by extracting image features using a MobileNetV2 model pre-trained on over one million ImageNet images, then adding and training a fully connected layer with the dataset obtained to classify the skin into one of the four types. For a ubiquitous and user-friendly use of the presented model, it was deployed on an android app through TensorFlow Lite. The main functions of the mobile application are: (1) accurately detect the abnormality based on the surface of the skin scanned by the user. (2) Suggest treatment based on the abnormality detected. (3) List nearby hospitals for treatment.

• Paper 2: A Garbage Profiling System Using Mask R-CNN Algorithm

  IEEE Xplore, International Conference for Advancement in Technology

  In this paper, a model developed using Mask R-CNN to aid the garbage profiling process has been presented. This model identifies plastic waste from garbage, following which data analysis is done to present the cities with high plastic usage levels. Using this data, the government will know which cities must take action to handle the usage of plastic materials. Today, even when separate bins are present for all types of waste, we have mixed waste in the majority of the containers. If debris is not segregated where it is generated, it won’t be recycled and may end up being in the landfill for a considerable amount of time, causing several problems. The dataset used for this study was obtained from Kaggle and a few external sources. After cleaning and labeling the data, the model was trained using the Mask R-CNN algorithm to classify plastic waste. It was deployed on a website to make the model user-friendly and visualized the results using Python’s Pygal library. The main functions of the website are: (1) Present areas where plastic usage is high, and segregation is not happening correctly (2) Present the amount of plastic usage in a given time frame. (3) Given an input image, generate the amount of plastic and non-plastic content in the image. Keywords— Mask R-CNN, Garbage profiling, Pygal, Flask, TensorFlow

Resume

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';