Home Point Financial
Profile and Alert Engine
- Built internal tool to identify hidden data relationships, dependencies, and quality issues
- Utilized Entity Framework to create relational database using .NET-based programming concepts and integrated Azure Active Directory for authentication and authorization
- Developed a web application using Angular integrated with Python-based microservices, and set up alerting system through Ops Genie and SendGrid
Propensity Universe
- Created a Mono repo with automated testing and deployment pipelines to streamline the deployment process and reduce the risk of errors
- Collaborated with data scientists to define deployment strategies, and incorporated ETL-based design in conjunction with PySpark
- Setup ML Flow model registry and Data Dog for logging and metrics collection to ensure models met performance and reliability requirements
Enterprise Deployments
- Designed and implemented enterprise-level feature store using open-source libraries to promote standardization, governance and ensure seamless access to feature data
- Created helm charts to deploy Airflow for data-driven teams to improve workflow management
- Worked closely with data engineering and setup Data Build Tool to modularize SQL code for maximum efficiency and maintainability with automated testing, code review, & version control
Automation
- Created an automated pipeline to deploy Docker images using a mono repo into Kubernetes
- Configured Selenium to interact with websites, including those with dynamic content and login requirements, and automated custom Python scripts to extract & parse the data
- Developed fault-tolerant pricing system using .NET, Python & used Redis for caching
- Pursued Domain-Driven Design approach to develop modular, extensible architecture to deploy and extend internal libraries in Python/Angular/.NET
~~~~~~~
Insight Data Science
Get Set Hedge
This project involved the development of a robust framework for factor construction, aiming to identify stock exposure to corresponding Exchange-Traded Funds (ETFs). By leveraging this framework, I sought to improve portfolio outcomes, reduce volatility, and enhance diversification.
To ensure flexibility and customization, I built a dynamic pipeline using the Gang of Four (GoF) design architecture. This allowed for seamless integration and adaptation of the framework to various investment strategies and market conditions. Additionally, I employed Docker for containerization, enabling efficient scalability and deployment on Amazon Web Services (AWS) infrastructure.
The "Get Set Hedge" project represents my commitment to leveraging advanced technologies and data-driven approaches in the financial domain. By combining machine learning techniques, domain expertise, and software engineering principles, I aimed to create a powerful tool for optimizing investment portfolios and mitigating risk.
~~~~~~~
Midmark Corporation
Stadiometer
With a strong focus on improving accessibility and accuracy, the Stadiometer project aims to provide a reliable and efficient solution for measuring the height of a standing subject. By leveraging the power of stereo cameras, we capture depth information and employ advanced computer vision techniques to accurately determine the distance to the subject.
Integrating machine learning models into the Stadiometer system takes height measurement to the next level. Through extensive training and data analysis, the machine learning algorithms are able to predict the height of a disabled person with remarkable precision. This innovation has the potential to significantly improve the lives of individuals with mobility challenges by offering a seamless and accurate height assessment method.
~~~~~~~
University of California - Irvine
Research
In the field of machine learning, data augmentation has proven to be a valuable tool for enhancing classifier performance, particularly for image-related data. However, I recognized a gap in the application of data augmentation to non-image limited data, which inspired me to delve into this area.
In my dissertation, I have developed two data augmentation policies specifically tailored for non-image limited data. These policies are designed based on the concept of measurement errors commonly encountered in scientific measurements. By inducing these errors, I aim to augment the limited data and provide a more diverse and robust dataset for cancer classification.
Another important aspect of my research is studying the role of Copy Number Variations (CNVs) in cancer manifestation. CNVs are structural variations that can provide valuable insights into the development of specific cancer types. To explore this further, I collected germ line data from The Cancer Genome Atlas Program (TCGA) and employed contemporary machine learning models to classify different types of cancer.
However, the performance of these classification models was hindered by the limited availability of data. To overcome this challenge, I integrated the data augmentation policies I developed into the workflow. The results were remarkable, as the performance of the classifiers experienced a significant improvement of approximately 3-4%. This improvement holds great promise for future research and analysis in the field of cancer classification.
By utilizing these data augmentation policies, my project not only enhances the accuracy of cancer classification but also facilitates the identification of dominant chromosomal regions that correlate with specific cancer types. This invaluable information can provide the medical community with deeper insights for further analysis and research.
Snake Bot
Using Genetic Algorithms, I designed a process that simulates natural evolution to train the Snake Bot. The algorithm starts with a population of randomly generated snake agents and progressively evolves them through generations. With each iteration, the fittest individuals are selected for reproduction, passing their genetic material to the next generation. This iterative process optimizes the bot's decision-making capabilities and enhances its gameplay strategy.
To further refine the Snake Bot's performance, I incorporated Reinforcement Learning techniques. Through trial and error, the bot learns from its actions and the resulting rewards or penalties it receives. By applying reinforcement learning algorithms such as Q-learning, the Snake Bot gradually improves its decision-making skills based on past experiences, ultimately achieving remarkable gameplay proficiency.
~~~~~~~
Shri G S Institute of Technology and Science
Sway Analysis
In this project, I collaborated with a team of fellow students to tackle the challenge of identifying and analyzing posture deformities. We designed a comprehensive approach that involved recording videos of individuals standing still for one minute, capturing their posture from various angles. The videos were carefully coordinated with 10 subjects, ensuring a diverse range of data for analysis.
Leveraging our expertise in image processing, we developed a sophisticated algorithm that processed the recorded videos and extracted key posture-related features. By applying advanced techniques, such as edge detection, contour analysis, and shape recognition, we were able to identify and differentiate various posture deformities accurately.
Heart Beat SensorThe heart beat sensor project allowed me to combine my technical skills in hardware design, data acquisition, and signal processing. I carefully selected and integrated the necessary components to capture heart rate signals effectively. Utilizing my expertise in embedded systems, I developed a streamlined and efficient solution to measure and analyze heart rate data in real-time.
Throughout the project, I explored innovative approaches and techniques to enhance the accuracy and reliability of the heart beat sensor. I paid particular attention to minimizing costs and reducing the overall size of the system, ensuring its practicality and usability in diverse applications.
~~~~~~~
Personal Projects
Sudoku Solver
This project showcases my passion for applying Artificial Intelligence and Constraint Satisfaction strategies to solve complex puzzles.
The Sudoku Solver AI is designed specifically to tackle diagonal Sudoku puzzles, which add an extra layer of challenge compared to traditional Sudoku grids. By leveraging advanced Constraint Satisfaction techniques, the agent efficiently solves these puzzles, providing a fascinating demonstration of problem-solving capabilities.
Pacman Agents
One of the exciting projects I worked on is developing a Pacman agent capable of finding food in an optimized way. To achieve this, I extensively tested various search algorithms such as Depth-First Search (DFS), Breadth-First Search (BFS), and A-star with varying heuristics.
By implementing these search algorithms, I aimed to enhance the efficiency and effectiveness of the Pacman agent's decision-making process. Through rigorous experimentation and fine-tuning, I sought to find the most suitable algorithm that would enable the Pacman agent to navigate the game environment intelligently and locate food with precision.
Adversarial Search Agents
In this endeavor, I delved into the world of adversarial search techniques to tackle challenging problems such as Knights Isolation and the N-Queens puzzle.
Using advanced algorithms and strategic thinking, I developed an intelligent agent capable of playing Knights Isolation, a game that involves maximizing movement possibilities while minimizing your opponent's options. Through meticulous exploration of search algorithms, including techniques like DFS, BFS, and A-star, my agent was able to analyze the game state, make informed decisions, and outmaneuver its adversary.
Cloak GUI
With the power of Python, OpenCV for Image Processing, and Tkinter for GUI, I've created a captivating application that brings the magic of an invisibility cloak to life.
Imagine being able to create your own invisibility cloak effect in a video. Well, that's exactly what this project allows you to do. The Graphical User Interface (GUI) provides a user-friendly platform where you can effortlessly apply the cloak effect to any video footage.
Facebook Page Post Scheduler
One of my notable projects is the development of a JAVA application designed to streamline the process of scheduling posts on Facebook pages. With the goal of increasing efficiency and saving time, I integrated a web crawler into the application. This powerful feature automates the collection of relevant content, ensuring that your Facebook page remains active and engaging for your audience.
Clinic Management Application
In this project, I leveraged my expertise in Java programming and database management to create a robust management application. By integrating MySQL, I empowered the system to handle and process large datasets efficiently, ensuring seamless operation even with substantial amounts of data.