Ashwin Shetty
Los Angeles, California adjnsb@r.postjobfree.com 213-***-**** LinkedIn Github
EDUCATION
University of Southern California, Los Angeles August 2019 - Present Master’s – Electrical and Computer Engineering (GPA- 3.95) Courses: Linear Algebra, Probability, Deep Learning, Mathematical Pattern Recognition, Machine Learning for Data Informatics, Advanced Computer Vision, Analysis of Algorithms
D.J Sanghvi College of Engineering, Mumbai May 2015 – July 2019 Bachelor’s in engineering – Electronics and Telecommunication Engineering (GPA – 3.83) SKILLS
AWS, GCP
Python, C++, Java
PySpark, Pandas, Scikit-Learn
Django, HTML, CSS, JavaScript
Cassandra, MongoDB, SQL
Tensorflow, Pytorch
PROFESSIONAL EXPERIENCE
Stylebot – Los Angeles June 2020 – Present
Machine Learning
Developed and deployed a Rasa chatbot over slack.
Assisted developers with migration from Dialogflow to Rasa open source.
Analyzed and identified issues with deployment over multiple slack workspaces to enhance user experience and determine expedient improvements.
Developed bot vocabulary by tracking user-bot conversations.
Wrote technical documents and manuals for software that was being developed. KAK Infotech Solutions – Mumbai June 2018 – July 2018 Software Development
Developed Machine Learning based solutions for tech companies.
Built a MATLAB program to detect currency denominations and forgeries using neural networks.
Assisted project heads in building and optimizing a chat bot in Python using Tensorflow PROJECTS
Real/Fake Disaster Tweet Detection Github View App
Performed EDA to understand general statistics about fake and real tweets such as length, word frequency, etc.
Preprocessed data by removing stop words, punctuations, numbers and tokenized the cleaned data.
Used term-frequency inverse document frequency transforms to convert text data to numerical features.
Trained a Logistic Regressor, SVM and Multinomial Naive Bayes model and achieved 80.79%, 78.24%, 77.21% val accuracy
Finally used a pretrained BERT (large-uncased) model to obtain 84.31% validation accuracy. Spoken Language Identification using LSTMs Github View App
Crowdsourced audio clips in English, Hindi and Mandarin
Removed periods of silence and extracted 64 dimensional MFC coefficients from audio to serve as features for classification
Implemented a Gated Recurring Unit and trained it to classify languages with an accuracy score of 94% in case English, 61.53% in case of Hindi and 70.59% in case of Mandarin CNNs for Image Colorization Github
Extracted bird image data from the cifar10 dataset and converted them to grayscale to serve as model input
Implemented K means algorithm using 32 clusters to find the main 32 colors components in the dataset
Converted the R,G,B component of images to the nearest main color component
Trained a deep CNN and a trailing dense network with dropout rate of 0.2 in the penultimate layer.
Achieved a final training accuracy rate of 85.56% and validation accuracy rate of 83.2%