Akhilesh Harish Mahajan
I have completed Master's in Data Science from Rutgers University. Apart from developing and deploying software products, I have experience of applying the predictive modeling techniques and statistical methods to real-world data and generate accurate predictions.
Data Scientist/Machine learning Engineer
NJ, USA
[email protected]
Work Experience
Tuutkia - Data Science Intern Sept 2019 ~ Present
Technologies : Python Flask, nltk
- Designed algorithm for detecting profanity and emojis in the description text using traditional Machine learning algorithms.
- Added APIs for handling profanity and emojis if encountered.
Rutgers WINLAB - Data Science Research Intern May 2018 ~ Nov 2018
Technologies : Python, R, Code Composer (Micro-controller programming)
- Designed and conducted experiments with ammonia sensors for data generation.
- Aggregated data from NIH, Winlab and Cornell and cleaned them using pandas.
- Proved that sensors are sensitive to frequency of sensing, temperature and humidity using Pearson's correlation coefficient which reduced electricity consumption in sensors by 90%.
- Developed an application to detect outliers in humidity senors, which filtered 99% of outliers in real-time.
- Developed R Shiny app to aggregate data from various sources and display time series plot of desired sensor ID for analysis.
Fiorano Softwares - Software Engineer Jun 2016 ~ Jun 2017
Technologies : Java7, gwt- Developed features on top of our product which reduced the log file size leading to a direct reduction in database hosting fees.
- Maintained enterprise server, which included refactoring of dashboard and JMS environment based back-end.
Nvidia Graphics - Software Engineer Jun 2015 ~ Dec 2015
Technologies : C++, C, CUDA, Java- Developed a library of CUDA applications to highlight the difference in performance in CPU and GPU, as well as between Maxwell and Kepler architecture.
- Added enhancements to CUDA profiler, which resulted in straight reduction in GPU program analysis time and improved their run-time efficiency.
Education
Rutgers University - M.S in Computer Science May 2019
CGPA - 3.6 / 4
BITS Pilani - B.S in Computer Science August 2016
CGPA - 7.4 / 10
Skills
Languages
- Data Science - Python, R
- Object Oriented - Java, C++
- SQL
- MATLAB
- Web - HTML, Javascript
- C
Libraries
Databases/Data store
- PostgreSQL
- Hive
- HDFS
- Pig
ML models
- Linear regression
- Logistic regression
- kNN
- Decision Trees
- SVM
- Graphical models
ML tools
- Error optimization - Gradient and co-ordinate descent, Penalty and barrier methods
- Feature selection - Mutual Information, PCA, Elastic-net, Chi-square
- Hyper-parameter tuning- Grid and random Search
- Kernel methods - Gaussian, Polynomial, Laplacian
Big Data
- Spark
- MapReduce
- Sqoop
- Flume
- Python Dask
Visualization
- Python - matplotlib, seaborn
- R - ggplot2
- Excel
Projects
Image Colorizer - Python
- Implemented k-Means algorithm to cluster the pixel colors of a colored image during training phase.
- Developed linear and logistic regression models to colorize the gray-scale image provided as input, which gave visual results nearly as good as ones colorized by neural network.
Imputation of Missing values
- Identified outliers and validated normality assumption using ggplot.
- Used feature selection to prevent overfitting.
- Developed Linear and Logistic regression, kNN and decision trees for regression and classification.
- Evaluated the models based on R-squared, identity error and odds ratio.
- Performed grid-search and random-search for hyper-tuning the parameters of kNN and decision trees.
The Art of Seeing - Flask, MySQL
- Moire pattern - Patterns developed by super-position of 2 identical patterns on each other.
- Developed a web-app which integrates a book on Moire pattern, lets the user understand the reason behind Moire effects and implemented a Moire image encryption algorithm.
k-Means in MapReduce
- Constructed an algorithm in Java to iteratively compute the clusters using the mapper and reducer paradigm with the help of job chaining.
Maze-runner - Python
- Constructed and solved mazes using BFS, DFS and A* algorithm.
- Generated harder mazes using the genetic algorithm.
