Projects
3D Localization using QR Codes - Robotics Intern
Developed a ROS-integrated Gazebo simulation for 3D QR code localization on moving objects. Designed custom Gazebo plugins and service wrappers to enable dynamic object control and motion simulation. Implemented homography-based QR code detection for tracking object states and relative motion. Integrated FastAPI for efficient QR code metadata management and real-time, service-based updates. Matsya - Autonomous Underwater vehicle
Worked in the software subdivision of a student team at IITB designing an autonomous underwater vehicle, Matsya, which is capable of navigation, smart decision-making and object detection enabling it to autonomously perform complex tasks in marine environments. Achieved 4th place globally in the finals of RoboSub 2024 Autonomy Challenge and 1st place in Technical Design Report among 41 teams from 7 countries in Irvine, California. Researched image processing and pose estimation techniques such as Otsu's thresholding and SIFT to improve robustness. Implemented an Action-based Navigator in ROS2 to add interrupt ability, leading to more efficient task scheduling. Integrated codebase into a Unity environment using C#, enabling parallel development and continuous testing of ideas. Mel-Frequency Vocal Classifier
Optimized traditional machine learning models for classifying songs by singer and category using Mel-Frequency Cepstral Coefficients (MFCCs). Leveraged statistical features of lower-order MFCCs to achieve an F1 score of 0.88. Automated data collection using YouTube APIs, gathering over 40GB of data and accelerating the process by more than 10 times. ASC Course Extractor
Developed a web scraper to extract the list of courses for each department in the upcoming semester from the ASC Portal, along with their restrictions, into a spreadsheet. Integrated a Twilio-based grade notifier to alert users when grades are updated on the portal. Converting data into a spreadsheet enables filtering by course time slots, making it easier to check for scheduling conflicts. This results in over 50% deduction in the time required to browse and select courses. Used Selenium, BeautifulSoup and openpyxl to achieve this. Vision-Based Sudoku Solver
Developed a CNN-based image detection model to identify digits, achieving 99.7% accuracy using a custom dataset and image preprocessing with contour detection techniques. Implemented backtracking for efficient Sudoku solving. Enhanced mathematical understanding of neural networks, focusing on optimization techniques, activation functions, and back propagation by implementing a CNN Model from scratch. Gesture Controlled Bot
Competed among 200+ teams as part of a four-member team to design, build, and navigate a gesture-controlled bot through an obstacle-filled track with maximum efficiency. Utilized an ESP32 microcontroller, MPU6050 for gesture-based spatial tracking, and L298N for differential speed control. Secured the Champion title by achieving the fastest track completion with bonus points. Algorithmic Trading - Literature Review
Studied Modern Portfolio Theory and Markowitz’s hypothesis, analyzing their impact on trading decisions. Gained hands-on experience in forecasting stock prices and volatility using statistical models like ARIMA, ARCH, and GARCH, as well as machine learning techniques such as LSTM for time series prediction. Explored various trading strategies, including Mean Reversion and Momentum Trading, to understand market dynamics and investment approaches. 