Pankaj Bhambhani

Currently involved in the Machine Learning for Data Science (MLDS) Lab, working on BirdCast, which uses CNN based image recognition to identify bird migration patterns from radar image data. We use WSRLIB, a MATLAB library for Weather Surveillance Radar written specifically for this task, coupled with a Convolutional Neural Network trained on a VGG-16 model.  Plans include scaling up the service to run on AWS clusters, and improve the performance of the CNN network.

Technologies – MATLAB, Python.

Worked as SDE for RummyCircle.com, one of the major products of Games24x7. Involved in Full Stack Web Development.

Technologies - Java, Spring, Spring Security, Jersey (JAX-RS), Hbase, Ehcache, Terracotta, Redis, Akka, MySql, Hibernate, MyBatis, Node.js. 

Following are the key projects I worked on during my tenure.

1. Analysis and Re-configuration of  Fund Manager Service

Worked with the Business Intelligence (BI) team and analyzed the data (using MySQL) containing the various real-valued transactions done by the player (cash deposit, cash withdrawal, playing game for cash) as well as system (service fee, bonus release,etc.) and recorded the discrepancies in the data, most of which were concurrency issues. Traced the root of these problems to the Fund Manager Service, which handled all user and system transactions using SQL Stored Procedures. Implemented a Locking Mechanism whereby the Stored Procedures updating the balance of an entity would obtain a row-level DB lock. This significantly reduced the number of discrepancies in the transaction data.

Technologies Used - MySQL (for analyzing the data as well as for Stored Procedures)

Team Size - 1

2.  Deposit Based Bonus 

As an extension of the existing Rummy Rewardz Store Program, where users could trade their reward points for items such as games bonuses and tournament tickets, we added a new functionality where the users could buy these items directly with cash deposits.

Role - Involved in the Controller and View related code changes in the Spring based application called RummyCircle (RC), which is the core user-facing service.

Technologies Used - Java, Spring, Hibernate, JavaScript

Team Size - 2

3. Fund Manager User Balance Bucketing 

This project involved separating the users' cash balance into two separate buckets (for marketing and legal reasons) - a deposit balance, where the user can deposit cash and use it to join games (but cannot withdraw from it) and a withdrawal buckets, which contain the users winnings and can be withdrawn any time. There were functionality changes in the all the various services using or updating a player's balance.

Role - Involved in code changes for  RummyCircle (RC) - which displays a user's transaction history, Fund Manager(FM) - which handles the user's transactions and Player Service Portal (PSP) - an internal web application containing used by the Customer Service and Risk Mitigation Team to help users and monitor fair play activities.

Technologies Used - Spring, Hibernate, MySQL (Stored Procedures)

Team Size - 3

4. Event Detection System (Extension) - 

Event Detection System E was a service that captures various user events (e.g. login/logout, deposit cash, playing a game) which help directing effective marketing campaigns and improve player retention. It internally used HBase to stored all its data and analyze it. Our task was to take this service out of beta stage, and implement it throughout the various services involved in gameplay.

Role - Involved in code change for EDS service as well as various other gameplay services, namely RummyCircle (RC) Desktop and Mobile Service, Tournament Engine (which handles rummy games and tournaments), Secure Payment (our Payment Gateway), Fund Manager, Player Service Portal, Tournament Service Portal (used to monitor the state of running tournaments).

Technologies used - HBase (for EDS), Java

Team Size - 2

5. Strengthening Application Security 

Based on the feedback received from an audit carried out by a third-party on request, we implemented various defense mechanisms to improve the security of our applications against malicious users and hackers. Security features were added to prevent attacks such as Cross Site scripting (XSS), Cross Site Request Forgery (CSRF), SQL Injection, Click-jacking, Session Hacking, Session Fixation, Web Parameter Tampering among others. Also, emphasis was laid on controlling the privileges available to users of internal applications which may contain sensitive user data, using LDAP (Lightweight Data Access Protocol).

Role - Implemented mechanisms for defense against XSS, CSRF, SQL Injection and Session Hijacking. This involved development as well as Penetration Testing. Also was a part of access control implementation for Player Service Portal, an internal app handling player related data.

Technologies Used - Java, Spring Security

Kali Linux PenTesting Tools Used - nmap, Wireshark, Burp Suite, sqlmap

Team Size - 3

6. Scaling and Optimization of Caching Services 

The existing caching services used Ehcache client as a cache to store non-persistent data for better performance. The data was synchronized across multiple services with the help of a Terracotta server. This mechanism had low fault-tolerance, increase in user traffic often resulted in crashes which meant the entire application system would need to be restarted. We worked on fine tuning the parameters of Ehcache clients so it would cache data only as long as needed and also limit the amount of memory it would use for the same. Also, we performed a horizontal scaling of distributed Terracotta servers so that it could easily handle a heavy load of 30,000 users (equivalent to a million requests) in peak times.

Technologies Used - Ehcache, Terracotta, Java

Team Size - 2

 
7. Optimization of Gameplay Bot

The existing gameplay bot, which was made two years before this project, could simulate gameplay for Cash Rummy Games and Rummy Tournaments by creating threads for users and games and following standard protocols (i.e. doing legit player moves). Owing to the excess use of threads (there was one thread per user per game), it could only handle a load of about 2000 users. The launch of bigger and bigger tournaments (the latest being 20k user) and the ever-increasing necessity to perform benchmark testing was the primary reason for this project. We carried out this optimization in several phases (I was involved in one of the phases). Initially, we optimized the necessary number of threads and could successfully handle upto 10k users. Later, we move to message-driven architectures like AKKA which significantly increased the user capacity of gameplay bot.

Technologies Used, Java, AKKA

Team Size - 5

8. Scalable REST-Based Non-Financial Service

Designed and implemented a new REST-Based service which could handle incentives that were "non-financial", i.e. other than real cash - for e.g., tickets, reward points, etc. As we were steadily moving towards a Service-Oriented Architecture, this service was designed to be stateless from the start and was scalable, achieving a benchmark of approx. a million simultaneous requests per node. The REST calls were handled using Jersey and the DB interactions were modeled using MyBatis.

Role - I worked on the part of the service that handles requests related to tournament tickets.

Technologies Used - Jersey (JAX-RS), MyBatis

Team Size - 2

Integrated features of Facebook into KDE’s micro blogging client, Choqok (Under the program Google Summer of Code 2012).  Using KDE's libkfbapi, on which I worked the previous year, we added the ability to view posts, like and comment on them and also create a new post.

Technology used: KDE4 based on Qt (C++ based GUI Framework).

Link - https://github.com/pankajb64/choqok-facebook

Build base KDE library, namely KFbAPI, which formed the ground for Integrating Facebook Features for Choqok (which was done next summer under Google Summer of Code 2012). 

Technology used: KDE4 based on Qt (C++ based GUI Framework).

Worked as an Automation Engineer with Merchandise Operations Management (MOM) Team at Oracle Retail on products like RMS (Retail Merchandising System) and Allocation. Used technologies like QTP (Quick Test Professional) for RMS and Selenium Webdriver for scripting and execution of test cases.