Hi, I am Shrikant Kendre

MSCS at University of San Francisco

a picture of me smiling

My Projects

Build a OCR model and multi-lingual custom NER model using CNN/ RNN with LSTM units

Researched on Techiques to improve previously used OCR in PhlexEview and performing custom NER Tagging across different languages. Developed a custom OCR model using CNN and Bi-Directional RNNs with LSTM units and calibrated against Connectionist Temporal Classification(CTC) loss. Deployed model using Azure Container Registry on Azure.

Improve the Ad Click Prediction Accuracy for Search Engine

Reseached on Various NLP algorithms for information. Implemented and Evaluated CDSSM(Convolutional Deep Structural Semantic Matching) over AUC, Log-loss reduction, Calibration plots. Worked on model deployment using Tensorflow serving with protocol buffers and gRPC.

Medicare Analytics using Health Outcome Surveys and Clinic Dataset

Developed analytical and ML model that assist Customers Marketing team by targeting right and influential population to improve the star rating of medicare plans by addressing specific measures. Used HOS data, clinical data, prescription data to train ensemble of models to calculate propensity for the members to have Urinary Incontinence.

Automatic Detection of Hateful text in Online content using Twitter

Built an intermediate module between user and twitter platform that monitor tweets made by the people. In the case of hateful/offensive content, it prevents the user from tweeting, hence maintaining and monitoring decency of the Platform. While doing the task also wrote a research paper for the same and is availabe here.

See my Work

Who I am

Pursuing Master's in Computer Science at USF

ML/ DL practitioner with exposure to full-satck development along with industry experience of 3 years, have strong knowledge of various ML/DL algorithms and statistical tools. Interested in applied Machine Learning to solve real world problems related to Computer Vision and NLP.

Along with my professinal interest in ML/DL I also find myself enhancing my software development skills by working on various personal projects. One of my recent porject encompasses developing a Ticket purchasing web applicatoin like EventBrite, that uses 2-tier model with front-end designed using Java(Jetty/ Servlets) and SQL(Azure database for MySql) as the backed and hosted the web app using App Services from Azure.

Shri's image again

My Work

Ticket Purchasing web app for Hosting/ Purchasing tickets to Events

Built a ticket purchasing web app like EventBrite. Designed a 2 tier- web application with Java(Jetty/ Servlets) along with HTML and CSS with JS as the front-end and SQL(Azure database for SQL) as the backend and hosted the application using App Services on Azure.

Handwritten Text Recognition using TensorFlow

Performed OCR on handwritten text with an accuracy of 80% on handwritten english text. Built a model using tensorflow(keras) trained on IAM dataset and evaluated against custom Connectionist Temporal Classification(CTC) loss, deployed solution as a web application using CherryPy.

Santander Customer Satisfaction(Kaggle)

Identified dissatisfied customers in the early phase and by doing so would allow Santander to take proactive steps to improve a customer's happiness before churning. Performed various feature selection methods including Filter Methods, Wrapper Methods and Embedded Methods. Model is developed using Random Forest/ Logistic Regression and evaluated on Accuracy(Train: 80%, Test: 81%) and Confusion matrix.

Image Classifier using Deep Learning

Worked on Deep CNN Model to build a classifier to identify the images(belonging to specific class), trained on CIFAR dataset. Built a Flask app to consume the model and deployed the app using docker container. Evaluated the performance based on user uploaded images and their feedback.

Automatic Detection of Hateful text in Online Content using Twitter API

Built an intermediate module between user and twitter platform that monitor tweets made by the people. In the case of hateful/offensive content, it prevents the user from tweeting, hence maintaining and monitoring decency of the Platform.

Given Lower Back Pain Symptoms, Categorize Abnormal/ Normal

Built a Binary Classifier using DNN to predict given a list of symptoms whether a person suffers from back pain or not. The data comprised of nearly 13 attributes and 310 observations, the DNN is optimised using RMSprop and evaluated on BinaryCrossEntropy, the metrices taken under consideration were Precision and Recall.

Regenerate Alphabets after Performing Dimensionaliy Reduction via AutoEncoders

Performed Dimensionality reduction to reduce the dimensions in the given input data and then regenerate the input data using this newer dimensional Dataset. The dataset consisted of 26 folders(A-Z) containing handwritten images in size 28x28 pixels, and each image is a gray-scale image. Autoencoders were implemented in 2 ways first via Dense NN and later via CNN(Encoding and Decoding), and was evaluated based on how perfeclty the input data is reconstructed.

Similar Document Retrieval via Wikipedia Dataset

Built a model that could return similar documents to the given document based on the input text as the features, dataset consisted of 3 columns namely URI, NAME and TEXT with nearly 50,000 observatoins. The KNeighborsClassifier algorithm is used to identify the similar document and the model predicts the name of the people closely related to the input document.