Projects
Multi-Variate Time Series Forecasting in Hydrology
(May 2024- Now) This aimed to measure the contribution of static variables on multivariate time series ML models using hydrology datasets.
- Technical Stack: python
Defining Affine Using a Functional Programming Approach
(Jan 2024- May 2024) Formulated a mathematical structure for the functionalities in Affine Space and implemented them using a functional programming and theorem proving language.
- Technical Stack: Lean
Emotion Fluctuation Identification In conversations
(Aug 2023- Dec 2023) Implemented a transformer-based NLP model to identify emotions through a conversation and utilize that to measure mental progress.
- Technical Stack: python
Privacy Preserving Machine Learning
(Mar 2023- Aug 2023)
This work is done with the focus on federated learning and improving the scalability and efficiency of personalized federated learning. I am studying several applications in federated learning including image processing and NLP which have higher privacy concerns as well as efficiency.
- Technical Stack: python, Git
Evaluating Alias Analysis methods
(Jan 2023- May 2023)
This project aimed to study the scope of alias analysis methods in LLVM and its impact on performance of overall compiler optimization. I experimented with various factors including considered application (program), level of the optimization and the implementation of AA method and through extending that I could show combining alias analysis methods improved the precision. In terms of execution time, the optimality of alias methods became insignificant. Advanced optimizations provided a 10% performance gain compared to basic alias analysis, but optimal alias analysis was not crucial due to negligible variance.
- Technical Stack: C++, Bash
Hyper Parameter Evaluation
(Aug 2022- Dec 2022)
The project was to study how the hyper parameters effect on the different machine learning algorithms. A range of algorithms where studied from logistic regression, SVM to convolutional neural networks. Intention of the study was to evaluate the tradeoff between bias and variance through accuracies and loss for the considered algorithms against different parameters.
- Technical Stack: python
Novatti Payment Gateway
(June 2021- Aug 2022)
Working on Novatti payment gateway which is the legacy product of Novatti acquiring. As one of the early members of Novati Acquiring, I have played a full-stack role contributing to every key component of the payment gateway which followed a serverless architecture. I worked on the implementation of several principal components including reconciliation flow, disbursement flow and merchant onboarding flow. Moreover, I actively contributed to the end to end implementation of PayXCrypto, which is the secure and compliant real-time crypto payment system of Novatti.
- AWS Stack: Lambda, DynamoDb, Api Gateway, SQS, SNS, Secret Manager, App Config, S3, CloudFront
- Other Stack: Java, Spring Boot, Javascript, Maven, React, GraalVM, PHP, Git, Bash
Conversation Based Indoor Localization
(May 2020- Dec 2020)
The aim of the system was to address the low accuracy problem of indoor positioning. Our proposition was to develop a chatbot based system where first the user is localized through the landmark details provided by the user. In the second phase, the user is navigated to the destination through continuous instructions. Within the development problems such as entity resolution, the ambiguity of landmarks and flawed data had been solved using various approaches and the effectiveness has been measured from a simulation at each step.
- Technical Stack: Python, Docker, Git
Driver Guidance System
(Oct 2019- Apr 2020)
Based on our studies on the taxi industry in Tokyo, a taxi guidance system was built for drivers in Tokyo, Japan. Real-time data analytic and large-scale multi-agent optimization is done for suggesting a driver with a recommendation. The complete system consists of two main components. One is the recommendation engine which considers both current vehicle supply historical demand and provides each vehicle with a suggested region. And the other one is the Point of Interest(POI) suggestion where POIs with the lowest queue time for hire (upon realtime data) in the suggested region are offered to the drivers. The implementation has been able to reduce the roaming time for taxis by a significant percentage.
- Technical Stack: Java, Docker, Git, Bash, SQL
Sysco Shop Ordering Platform
(Feb 2019- Oct 2019)
In the beginning, I implemented a complete prototype of a point of sale system to get a broad understanding about the infrastructure where technologies including React, NodeJs and MongoDb were used. Then as a member of a development team in a World’s largest food services company, I worked with a team of 8 engineers that managed some of the core services in the Ordering platform where I involved in developing features required for the platform and worked on fixing bugs reported within the system. In fact, as a software engineerI had to work on a range of tech stack including Java, Spring Boot, SQL and dockerfor the backend development and React for the bug fixes needed in the front end.
- Technical Stack: Java, Springboot, React, Javascript, NodeJS, Git, Bash, SQL
Vehicle Clustering system for Highways
(Jan 2018- Dec 2018)
Final year project was a simulation study for development of dynamic vehicle clustering algorithms for highways, where each cluster is driven by the leading vehicle.The vehicles where vigorously swapped among clusters to increase the convenience of drivers while minimizing total travel time.The simulation has been done on omnet++ framework where vehicles where simulated as independent nodes which are capable of communicating with each other.Design was implemented using C++. Two algorithms were developed where one is centralized and the other one is decentralized. We compared those implementations through pre-defined metrics.
- Technical Stack: C++, Omnetpp
Internet Traffic Shaper
(June 2017- Dec 2017)
This is a internet traffic shaper implementation, which shares available internet bandwidth fairly within a hierarchical network ,upon preconfigured min.max and weight values while maintaining suitable QoS for each tcp connection.All of this has been attained by calculating a sheduel time for each packet.Though this requirement has been tried to addressed through ”Hierarchical fair service curve algorithm” previously, traffic shaping was not smooth and failed for higher number of connections.So a modern algorithm was designed which dynamically shared bandwith through the hierarchy.This was supported with another development to keep track of active/inactive state of each connection.The project was designed on software domain using C++.This was also implemented on FPGA and total calculation time could have been significantly reduced through a parallel architecture.
- Technical Stack: C++, Git, Bash, Verilog, FPGA
Hardware Accelerator for CRC Calculation
(June 2017- Dec 2017)
This is a project done for a local competition for fast CRC calculation which marked the best time in the competition.The implementation was a hardware acceleration through FPGA. A new parallel algorithm was designed which generates an advanced look up table for a given CRC key.Hardware accelerator has been used as a slave of a processor, so processor could hand-over entire calculation when there is a need of CRC calculation.Coded using verilog.
- Technical Stack: Verilog, FPGA
Maze solving and colour arrow following robot
(July 2016- Aug 2016)
The robo was implemented for a robo competition(Robofest - SLIIT 2016) where first the robot explores and maps a maze.Then robot arrives to a given location through the shortest path. After The robot exits it grabs a box and follows set of arrows using image processing and place the box.Arduino and Raspberry Pi platforms were used for the development.
- Technical Stack: Arduino, Robotics,