I'm

• Reverse-engineered the pathing behaviors of the ghosts in MS. PAC-MAN (1980) for an autonomous robot system using ROS.
• Developed drivers to integrate key components like LIDAR, enabling real-time navigation and path planning for the MiRo-E robot.
• Implemented a custom D* algorithm to control the ghosts' movement, achieving a 43% success rate in defeating Pac-Man in a physical reconstruction of the game’s first level.

• Engineered a customizable boot time estimator for Xilinx's Versal ACAP, enabling users to build tailored boot images based on specific requirements and target boot times.
• Reduced average boot time by 30% through optimized image generation, saving an estimated 7,600 hours annually across the product line.
• Leveraged SQL and DBT to clean and transform large datasets of Versal ACAP boot time information, collaborating with San Jose teams for data acquisition.
• Integrated the estimator with Django, ensuring scalability and accessibility for developers and end-users.
• Provided on-call support to resolve critical boot security issues, contributing to the overall stability and security of Versal ACAP product line.

• Developed a Integrated Peripheral Surveillance System (IPSS) utilizing LIDAR, cameras, and microwave sensors to enhance security at coal mine entrances by restricting unauthorized access.
• Designed and implemented machine learning pipelines incorporating ARIMA time series forecasting models to predict gas composition levels in coal mines, crucial for maintaining safety standards.
• Successfully improved gas composition prediction accuracy by 12% compared to previous methods, directly contributing to enhanced mine safety protocols and risk management.

➔ Engineered end-to-end data pipeline processing 100,000+ Steam game entries.
➔ Utilized AWS EC2 for scraping, S3 for storage, Polars for ETL, and PostgreSQL for data management.
➔ Implemented Grafana dashboards for analytics and AWS Lambda for automated data updates.
➔ Containerized entire pipeline with Docker.

• Addressed the challenge of accurately representing MWEs in NLP.
• The improvements proposed for the BERT model include a batch normalization layer for word vector quality, a loss function constant for better sentence vectors, and a channel attention mechanism for refined word vectors.
• Evaluated the enhancements using a dataset of English and Portuguese MWEs and achieved improved scores in MWE representation tasks.

• Utilized PySpark and Hadoop for large-scale (10GB+) dataset analysis.
• Developed liability claim prediction models using Linear and Logistic Regression.
• Contributed to movie recommendation systems (Collaborative Filtering) for user-customized suggestions.

• Designed and developed a 3-D particle effects rendering system.
• Leveraged CUDA to enhance rendering performance by nearly 95%.
• Implemented algorithms for realistic and dynamic visual effects.
• Demonstrated the project's capabilities to peers and faculty.

Studied the construction of coherent states and the representations of said states using Group Theory and its applications.

Implemented Localization (GridMap and EKF Localization) and used OpenCV alongside ROS to create a fully autonomous rover. Conducted field experiments with a prototype to achieve tennis ball tracking and retrieval. Successfully cleared Indian Rover Challenge (IRC) 2022.