Overall 4+ years of experience as a Software Tester and 1-2+ year of hands on project experience in Natural language Processing, Data Science and Computer Vision. Presently an AI Research Fellow at Fellowship.ai
Generative Models, Optimization, Reinforcement Learning, Unsupervised Learning
TOOLS AND TECHNIQUES
Python, R, C, C++, Java, Numpy, Scipy, Matplotlib, Seaborn, Scikit-learn, Keras, Pytorch, Tensorflow, Fastai,
CNN, RNN, familiarity with OpenCV, Tokenization, parts of speech tagging, stemming, lemmatization, named
entity recognition, TF-IDF, topic modeling, bag of words, word vectors, seq2seq, LSTMs, Transformers,
knowledge of semantics, syntax, morphology, phonology, SQL, familiarity with MongoDB, functional
programming, OOP/OOD, familiarity with CUDA, distributed computing – multi GPU,
NLP Use Cases- Text Classification, Sentiment Analysis, Language Modelling, Image captioning, entity
linking, relation extraction, representation learning, knowledge graph
Data Visualization: Tableau, matplotlib, seaborn, ggplot
Machine Learning Algorithms: Linear Regression, Logistic Regression, Random Forest, KNN, K-means
QA: Familiarity with Selenium and QTP, HP QC, Team track, JIRA, SharePoint, Excel, MATLAB
Version control: Git
Cloud: AWS, Google Cloud (GCP)
Methodology: Agile, TDD, BDD
FELLOWSHIP.AI, Location –Irving, TX, Remote
ML/DL Research Fellow, Jan 2019– present
Oven-based-Raw-Food Image Classification
The AI-enabled smart ovens are on the rise. Along with built-in cameras, computer vision algorithms could propose a recommended temperature and time for different food types – making cooking more convenient for customers. The large variety of food types for oven and the increasing market demand for better smart ovens are driving up needs for identifying them more precisely through food image classification. Team (6-7 members) collected around 1k+ images for 30 classes- bagels, carrots etc.- through Google/Bing Images, around 7 classes/person. Achieved accuracy of around 89% on RESNET34 and 94% on RESNET50.
Burn Severity Image Classification (02/01/2020 – 02/29/2020)
Participated in the burn severity image classification of 4 classes- first degree, second degree, third degree and none. Used the platform.ai (https://platform.ai/) platform – a unique visual and code-free approach to labelling images and train deep learning models – to labels images based on the projections and trained the resent34 model on the labeled data.
Presented research paper with co-presenter (1/22/20)
Adversarial Examples Improve Image Recognition - to the weekly Reading Group (https://arxiv.org/abs/1911.09665)
Presented myself the following paper: Self-training with Noisy Student improves ImageNet classification (2/5/20) - https://arxiv.org/abs/1911.04252
Creating High Resolution Images with a Latent Adversarial Generator (due on 3/26/20) -https://arxiv.org/pdf/2003.02365.pdf
Watchman (Work in Progress), since (3/2/2020)
Watchman allows you to detect out-of-distribution data of your machine learning, computer vision product. The library is based on research into detecting out-of-distribution data undertaken at fellowship.ai. Presently developing an API as a service to detect out of distribution based on training of introspection net and in-distribution model on in-distribution datasets and out-of-distribution datasets.
TATA Consultancy Services Ltd., Gurgaon, INDIA
Software Tester, Dec 2012– Nov 2018, Clients – CVS Pharmacy, Grainger.com
Tested standalone application of CVS Caremark (Pharmacy) client.
Prepared test case scenarios, test cases from Business Requirement Documents (BRDs) and use cases for functional testing of the application.
Identified test cases for Regression Testing and various phases of testing.
Executed test cases, logged defects in HP QC/Serena Business Manager and verified completion of defect life cycle. Troubleshoot test issues, recorded test results, tracked and prioritized defects.
Worked on production defects testing slotted in the coming releases.
Delivered training sessions and presentations for cross-team co-ordination
Executed some test cases on QTP and developed sample scripts for Selenium
Word embeddings: Trained a word embedding on IMDB dataset and achieved a validation accuracy of around 88%. Visualize these embeddings using the Embedding Projector.
Text classification: Performed text classification on IMDB dataset using Bi-directional wrapper with an RNN layer and achieved an accuracy of around 85% on test dataset.
Text Generation: Generate text using a character-based RNN on a dataset of Shakespeare's writing from Andrej Karpathy's- The Unreasonable Effectiveness of Recurrent Neural Networks.
Machine Translation: Trains a sequence to sequence (seq2seq) model for Spanish to English translation on a language dataset provided by (http://www.manythings.org/anki/)
Image captioning: Downloads the MS-COCO dataset, preprocessed and cached a subset of images using Inception V3, trained an encoder-decoder model, and generated captions on new images using the trained model
Gone through the following MOOCS:
FASTAI practical deep learning course (Part1) - https://course.fast.ai/
Andrew Ng Deep Learning Specialization: https://www.deeplearning.ai/deep-learning-specialization/
Andrew Ng Machine Learning - https://www.coursera.org/learn/machine-learning
Applied AI with Deep Learning, IBM, Coursera
Quantized modulation diversity for 64-QAM, IEEE, Apr 2012
MS in Supply Chain Management degree, 2018 (STEM) program is approved by the internationally recognized Institute for Supply Management (ISM) that explores the key issues associated with the design and management of industrial supply chains, including methods for improving supply chain operations by lowering costs and improving quality.
Undertaken the following relevant subjects
OPRE 6301 Statistics and Data Analysis
OPRE 6302 Operations Management
OPRE 6366 Global Supply Chain Management
OPRE 6332 Spreadsheet Modeling and Analytics
OPRE 6398 Prescriptive Analytics
OPRE 6304 Operations Analytics