Hello and welcome to my portfolio!
I am a Software Engineer with 5+ years of experience. I spent the last 3 years of my career as a Machine Learning Engineer (Python), working on projects in the domain of Computer Vision and Natural Language Processing. Prior to that, I was a Software Developer(JAVA) and was employed by a product based company headquartered in Houston. Read on to find out more about my projects.

RECENT PROJECTS

Pet Breed Identification

Tech Used: Python, Keras, Flask, Nginx, AWS ec2 Linux instance
Domain: Computer Vision - Image classification
Dataset: Google images, Dogs and Cats Breeds Classification Oxford Dataset

The pet breed prediction model was implemented using Xception model (a variant of inception CNN model) in Keras. Transfer learning was used to populate the weights of the top layers of the model and only the bottom layers were newly trained using our dataset, thus saving time in training the model. The pre-trained weights were obtained by training the model on ImageNet Database, one of the largest and most diverse image datasets. The resulting trained model was able to predict more than 400 breeds of cats and dogs with an accuracy of more than 75%. The model was deployed on AWS and nginx was used to serve requests for prediction.

Fish Species Detection

Tech Used: Python, Tensorflow Object Detection API, Flask, Nginx
Domain: Computer Vision - Object Detection
Dataset: Images collected manually and annotated using LabelImg

The goal of this project was to identify species of fishes that are present in the image captured by the user. Tensorflow’s Object Detection API provides multiple model architectures that can be trained and inferences can be made using these trained models. For this problem, I chose Faster RCNN Inception model architecture and trained the model on the fish species dataset. Weights from training COCO dataset was taken as initial weights. The trained model was deployed on AWS and nginx was used to expose the API endpoint and serve prediction requests. The training session was setup on AWS Sagemaker during later stages of the project since it saved a lot of training time.

Extracting Text from Identity Card for Autofilling forms

Tech Used: Python, OpenCV, Pytesseract, Fuzzywuzzy Flask, Nginx
Domain: Optical Character Recognition (OCR)

The main idea behind this OCR model is to extract details from a scan of the identity card, map details on the card to fields on the form. OpenCV was used to pre-process the image, PyTesseract, to extract text from image and fuzzywuzzy, to find the closest match to the specific keywords we are looking for in the image

Backend Functionality Development of a Networking App for Entrepreneurs

Tech Used: Django, Nginx
Domain: Backend Development

Similar to other social media apps, every user in the system has a profile. Users can post text and images for others to view, create and attend events, post jobs and other updates. RESTful API’s for this app were implemented using Django and data was maintained in MySQL database.

Analysis and manipulation of data

Tech Used: Python, Pandas
Domain: Exploratory Data Analysis (EDA)
Dataset: Kaggle Superstore Dataset

Exploring, analysing and manipulating data in a dataset as part of pre-processing so that they can be used later for prediction. This project is explained in detail on my blog here.