Engineer Python

Noopur Koshta

******.******@*****.*** +91-626******* https://www.linkedin.com/in/noopur-koshta/ PROFESSIONAL EXPERIENCE

Persistent Systems Pune, India

Machine Learning Engineer Nov 2016 - Present

• Clinical Diagnosis – Extractive summarization of clinical reports to obtain the ICD codes out of the report summary via traditional LSTMs.

• Diabetic Retinopathy Detection – Approached the classification of retinal images into stages of diabetic retinopathy as 4-stage classification problem using CNN with attention mechanism. This was a proof-of-concept developed with an accuracy of 85%.

• Web Analytics Dashboard – Predicting membership renewal on the basis of various combinations of parameters selected via dashboard. Overall implementation was done using python library plotly, and was hosted as a Flask application.

• Anomaly Detection – Irregular spikes in ticketing system were analysed over the course of at least 5 years, along with the topmost categories of tickets which were raised was extracted via LDA (topic modelling technique).

• Optical Character Recognition (OCR) – Used OpenCV and PyTesseract to extract text from the images of receipts. Pre-processing steps included cropping on page layout, deskewing, shadow removal, and increasing contrast to get more accurate text output.

• Web Analytics Tool – Developed membership transition prediction solution based on ensemble classification methods on top of data processing pipeline developed using Spark. PERSONAL PROJECTS

Predicting Bike-Sharing Patterns

• Built and trained multi-layer neural networks from scratch to predict the number of bike- share users on a given day.

Dog Breed Classifier

• Given an image of a dog, the designed algorithm whose base was CNN, produced an estimate of the dog’s breed. If supplied an image of a human, it produced an estimate of the closest-resembling dog breed.

• Two models were developed and compared, one with CNN designed from scratch and the other with transfer learning employed (here the network used for transfer learning was VGG-19).

Generate TV Scripts

• Performed sentiment analysis and generated new text, using Long Short-Term Memory

(LSTM) Networks that resembles a training set of TV scripts.

• The generated script can vary in length, and looked structurally similar to the TV script in the original dataset.

Generate Faces

• In this project, DCGAN was defined and trained on a dataset of faces. The goal of this project is to get a generator network to generate new images of faces that look as realistic as possible.

Sentiment Analysis Model

• Used Amazon SageMaker to construct a complete project from end to end. The goal of this project is to have a simple web page which a user can use to enter a movie review. The web page will then send the review off to the deployed model on SageMaker, which will predict the sentiment of the entered review.

SKILLS

Programming Languages: Python, R, Bash

Frameworks and Libraries: PyTorch, Keras, Flask, Apache Spark, OpenCV, PyTesseract Databases: MySQL

Cloud Computing: Amazon EC2, Amazon SageMaker, AWS Lambda HONORS

• Secured 1

st

position in Workshop on Aptitude Development (WAD), an inter-college competition held in Nagpur, India.

• Received scholarship for securing 1

st

position in third and fourth semesters in B.E.

(Information Technology), along with being 10

th

Nagpur University Topper in third

semester.

CERTIFICATIONS

• Deep Learning Nanodegree, Udacity

• Machine Learning Nanodegree, Udacity

Contact this candidate