Machine learning and Reinforcement Learning projects (Python)
 Supervised learning – Used large two public datasets to do model complexity analysis and
learning curve analysis to do comparison between different supervised learning algorithms
Decision Tree, Neural network, Boosting, KNN, SVM
 Randomized optimization – estimated the weights of neural network used for public dataset
using randomized optimization algorithms Random hill climbing, Simulated annealing, Genetic
algorithm and Mimic and also implemented these algorithm’s to implement model comparison
for 2 NP hard problems – Knapsack and One max.

Unsupervised learning and dimensionality reduction – used clustering (Kmeans and Estimation
Maximization) methods to do model selection on two large public datasets and also performed
dimensionality reduction using PCA, ICA, Random projections and Ida and analyzed
the results of applying clustering on reduced datasets. Applied the clustering algorithms to
the same datasets treating the clusters as if they were new features and ran neural network
learner on the newly projected data and analyzed the results.
 Implemented Markov Decision process for Frozen Lake problem and Taxi v2 problem using
open Gym AI and compared performance using value iteration as well as policy iteration for
smaller size and larger size of the problem respectively and also solved the same problems
using model free Q learning algorithm.
 Replicated the results in Sutton’s paper (Learning to predict by Temporal difference methods)
random walk experiment
 Implemented Lunar Lander experiment using Open Gym AI package and proved convergence
using Deep Q Learning algorithm. Trained the Lunar lander agent to achieve an average score
of 200 over 100 consecutive episodes. Performed series of experiments using difference
range for model hyper parameters and observed the impact on the agent convergence.
 Replicated the results in RoboCup experiment described in Correlated Q-Learning by Amy
Greenwald and Keith Hall paper using difference variants of correlated Q learning , Friend Q
and Foe Q learning approaches.

Artificial Intelligence (Python) projects
 Implemented Game playing agent using minimax, alpha beta, iterative deepening and alpha
beta pruning approaches respectively. The goal was to win against random player 100% of
the time and trained player using minimax and alpha beta 70% of the time and iterative
deepening and alpha beta trained agent 60% of the time.
 Implemented various search algorithm (breath first search, uniform cost search, a-star, bidirectional
UCS, bidirectional a-star and tri-directional search methods) for a graph and performed
the performance analysis.
 Implemented Bayes net, decision tree, expectation maximization, hidden markov model
model from scratch using python.
Data Science projects executed (R Programming)
 Implemented data analysis for large movie dataset and implemented preprocessing (Data
standardization, binning, label encoding etc) using R and performed data visualizations on
feature’s to identify dependencies and patterns or trends.
 Performed linear regression and logistics regression on movie dataset and estimated the performance
of each algorithm
 Implemented time series and Geospatial data visualization’s in R.

Big Data projects(Scala, Hadoop, Hive, PIG, HBase, Spark, Pytorch)
 Implemented multiple projects for Health Care for large dataset performing predictive modeling
utilizing Scala, Hadoop (Map reduce), PIG (ETL), Spark (ETL processing) and Spark
graphx and MLib modules
 Chest X Ray diagnosis utilizing Deep Convolution neural network by using transfer learning
from DenseNet 121 layer pytorch CNN model. The model was trained using NIH Chest X Ray
3
dataset which contains 110K images from approx 38K patients. Overall RUC AOC of 0.8 was
achieved in this research.
Android development (Java)
 Developed a Encryptor utility app to produce a cipher text based on message, key number
and increment number as input. The core encryption logic was developed in java.
 Implemented a word game (simplified form of scrabble) for Android OS following Agile software
development methodology using Java. Created game manual, project plan, test plan,
design document, use case model, component diagram, deployment diagram, UML design,
low level design as part of project documentation.

Project Description
Sirius XM Satellite Radio is an American broadcasting company that provides three satellite radio and online radio
services operating in the United States: Sirius Satellite Radio, XM Satellite Radio, and Sirius XM. Supply chain
component in SiriusXM handles OEM, Retail, and SMS (Subscriber Management Systems) data flow and business
processing logic.
Highlights
 Developing rest api’s using swagger specs and implementation in Go and kafka for new services as well as
services as part of migration project from tibco to go lang.
 Developed Rest and SOAP API’s, implemented several asynchronous message processer interfaces for
complex business functionality and implemented large data batch process using file adapter in TIBCO.
 Developing batch process and maps and web services in Sterling integrator.
 Written complex bps and map in Sterling Integrator using map user exists, Java task service and command
line adapter utilizing shell script, python scripts.
 Performed GIS inline upgrade and fresh installation.
 Built scripts for auto deployer, complex event subevent validation map, environment comparison and edw
scripts in python utilizing modules like numpy and pandas.
 Process improvement initiatives, performance improvements, testing automation, documenting coding
standards, maintaining supply chain wiki. project status reports, assisting and training team members.

Limited Brands, Columbus, OH
Skills used: IBM Sterling Integrator (mapping EDI, XML, SAP IDOC, Flat files), Shell and pearl scripting
Project Description
Limited Brands is a parent company of brands like Victoria’s Secret, Pink, Bath & Body Works, C.O. Bigelow etc.
Limited Brands sells lingerie, personal care and beauty products. The logistics operations of business involve EDI
transactions between L Brands, Factories, forwarders, carriers, consolidators, brokers, customs, and end customers.
EDI also support bank transactions for vendor payments. The backend systems consist of SAP retail, GTS,
SCEM, Rockport, MANU, PKMS, JDA, TIBCO etc.
Project Responsibilities
 Working on Supply chain B2B projects -Project execution in each step in software development life cycle from
design, development, unit testing, and Integration testing, deployment and production support utilizing agile
development methodology.
 EDI partner setup and connectivity setup using AS2, developing EDI to EDI and EDI to application maps like
flat file and Idoc and writing business process for file transmission to backend systems.
 Creating EDI Implementation guides utilizing EDISIM, participate in project workshop with internal partners and
external partners in different phases of project implementation
 Document support procedures, outage procedures, gaps analysis, technical, functional and architectural
specifications.
 Identify system improvements and implement solutions to scale system performance, optimize business
process, map, and tuning process, and thread load balance.
 Refactored generic inbound and outbound business process for ansi x12 and edifact to achieve significant
performance improvement.
 Written perl script to identify EDI file separators which is invoked from GIS using command line adapter.
 Upgrade SI version and installation of patches

Ecolab Inc, St Paul, Minnesota (USA)
Skills used: Gentran Server (Linux), TIBCO Business works and Business connect (Windows), Mainframes
(Cobol)
Project Description
Ecolab is the global leader in cleaning, sanitizing, food safety and infection prevention product and services. Ecolab
has bought IBM Sterling Gentran and TIBCO Inc TIBCO Business work and Business connect for their EDI
solution. IBM EDI team role is to own, support and maintain EDI applications at Ecolab.
Project Responsibilities
 GIS / TIBCO System administrator role in performing system component upgrade, installation, server
maintenance, system backup, disk space management, certificate installation, system cleanup and archiving.
Taking pro-active steps in installing hot fixes and patches to ensure system is current with market
standard and pose low risk to external network threat. Administrative tasks also include access control to
new system users and defining and maintaining access roles. Version control of system components is
maintained using ChangeMan DS tool. ChangeMan DS system is a client tool that allows the administrator
to maintain multiple versions of code and defines a procedure to back-out to previous versions if necessary.
 Monitoring IBM Gentran system components, system interfaces and adaptors for checking system health
and performance. This requires taking pro-active steps to ensure system availability and up time. Steps
consist of advanced activities like system performance tuning, resolving resource conflicts, creating loadbalance
servers, fault tolerant machines and deadlocks resolution in real time environment.
 Complex error analysis and debugging of system hung up data, message process errors, communication
errors, compliance errors and translation errors to ensure business data integrity. Activities might include
complex data recovery of impacted files or trading partner communication in resolving data issues. The
tasks are accounted in problem management and change management utilizing Ecolab Remedy and Service
Now tool.
 Development and testing utilizing TIBCO designer component of TIBCO business works
 Adding new functionality in TIBCO projects and implementing new procedures for data communication.
6
 Setting up business trading partner and business agreement in TIBCO business connect tool utilizing protocols
like X12, Ezcomm or EDIFACT. Implementing partner communication scripts like FTP, HTTP or AS2,
maintaining EDI standard protocols transactions guidelines and defining smart routing transactions if necessary