Venkata Avinash
Hello there for detailed contact information please fill the contact form, I will get back to you as soon as possible.
If you are looking for a detailed resume have a look at my LinkedIn page
If you are looking for some sample codes I have written have a look at my Github
Education
Rutgers University – Masters in Information Technology & Analytics
Indian Institute of Technology Madras, INDIA – Bachelors & Masters of Technology in Chemical Engineering
Skills
Programming
Python, R Programming, Java, SQL, C, HTML, Google TensorFlow, Keras, Scikit-learn, Matlab, PySpark
Machine Learning
Experienced in Google TensorFlow, some of the models I worked on includes but not limited to Deep Neural Networks, Recurrent Neural Networks, Convolutional Neural Networks, Attention Models, Linear regression, Logistic Regression, SVM, Generative Learning Models, K-Means, KNN, Decision Tree, Ensemble methods, Random Forest, XGBoost, AdaBoost.
Experienced in feature engineering and optimization, well versed in Linear Algebra, Convex Optimization, Multivariate Calculus Continuous, discrete probability distributions, Bayesian Statistics, Software Engineering, Link Time Prediction and Imbalanced-data learning methods
BigData
Spark, Hadoop, Hive
Visualization
Tableau, D3.js, ggplot and matplotlib
Projects
Predictive Crime Mapping using RNN LSTM
Nov 2017 – Present
Developing a predictive crime mapping neural networks model using Tensorflow.
This Model uses data from existing criminal database and predicts likelihood of a future crime of a given person, which will used for intervention, reform and rehabilitation of criminals while preventing crime from happening.
Work Done
Developed new features capturing the time variance and age dependent charges to better capture the dependencies.
Built a deep fully connected neural network using new sampling and bagging methods on the data from previously built SVM models improving the overall predictive performance.
Extended the Deep network by using a combination of bagging and ensemble networks to improve the F1 scores.
Developed a framework to capture the best performing hyper parameters which will be used to improve the RNN models
Work in progress
Building a RNN-LSTM model with increased feature granularity using a new multi class classification approach.
-Worked on language processing for emotional classification of text data parsed from social networking websites as part of an employee evaluation system.
-used twitter streaming for building the dataset and bayesian classification method to predict the emotional state of the employees.
Spendze
Feb 2018 – Feb 2018
Designing a deep learning model to be able to learn and predict spending patterns of consumer based on bank transaction data. This model will be used along with a coupon and product manager to give users custom offers which will maximize the value of expenses.
Link time prediction model for buses using neural nets in R
Dec 2016 – May 2017
Developed a Neural Network model for link time prediction in R for chennai MTC buses as a part of Intelligent Transportation System project sponsored by Ministry of Urban Development.
Developed tools to visualize locations of all 350 buses using OSM in real-time. Also developed a module, which tracks the number of functioning devices and the quality of the data being sent. Both tools were developed using Python
Optimization of Acoustic Cleaning Parameters in Semiconductors Surfaces | Graduate Thesis
May 2014 – May 2015
Experimented on PMMA (poly methyl methacrylate) particles generated during the transfer of graphene to copper substrate
Quantified the cleaning efficiency of different process variables and Optimized the process by statistical inference drawn using Taguchi Methods
Effects of Agricultural Products on price of Oil in southeast asian countries
May 2013 – Jul 2013
Forecasted the price of oil using ARIMA and ARCH-GARCH modeling
Studied casual effects between the prices of Rice, Bran on crude oil using normal copulas and Bernoulli copulas and getting an empirical relation between the casual variable and dependent variable.
Attrition prediction in HR training data
May 2018 – Present
Developed a Fully connected neural network model using IBM HR data with accuracy of 88% and F1-score 0.71. This project is done in alliance with PRISM Compliance to replicate the model in predicting attrition in HR training data.
Other Projects
YOLO object detection – Deep learning
Real time object localization and classification based on research paper ”You Only Look Once: Unified, Real-Time Object Detection”
SQUAD Q&A – NLP
Built a model to answer questions based on the 500+ wikipedia articles
Computer vision – Deep learning
Image classification of monkey species